* fix(litellm): report zero cost and cache-hit attrs on cache-served spans
* fix(litellm): stamp cache-hit marker on HiddenParams carriers and restore probe side effects
gauntlet-mini on ace57d25f verified two live gaps. TextCompletionResponse and
ResponsesAPIResponse carry _hidden_params as a HiddenParams object, so the
dict-gated marker write skipped them and the served hit still stamped the full
price. The marker write is now container-agnostic, which it already was for the
partial-embedding stamp, and a regression test exercises the carrier end to end.
_uncached_response_cost also restores billed_token_rates, zero_cost_warned, and
the response_cost_failure_debug_information key alongside cost_breakdown, so the
probe leaves no pricing state on the call it reads.
* test(caching): cover cache-hit marker on HiddenParams carriers through the caching handler
* fix(logging): keep the uncached price probe's restores to the state it can leak
* fix(otel): emit V2 LLM-call span for cache hits with explicit cache attributes
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(otel): read cache payload fields defensively without casts
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* chore(otel): clarify pyright suppression reason on cache payload reads
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(otel): build the cache-hit view from typing_extensions.TypedDict so import works on Python 3.10 and 3.11
* fix(otel): report cache hits at zero cost everywhere and mark misses
Cache hits now carry a zero cost breakdown and a zero hidden response cost,
the /v1/messages hit cost header follows chat and responses hits, every span
carries litellm.cache_hit, V2 hit spans drop server.address, an async hit keeps
the provider-prefixed model name, and the V2 attribute budget counts preset keys
* fix(otel): send a 0 cost header on cache hits and keep hits out of deployment TPM
Cache hits now send x-litellm-response-cost: 0.0 on chat, responses and
/v1/messages, while a fresh call whose cost is 0 still sends no header.
The async hit path stamps the same model name as main again. Stamping the
provider-prefixed name made a hit's tokens count toward the deployment's
TPM limit, so repeated cached requests hit 429s under usage-based routing.
Drop comments that restated fields and annotate new test values as Final.
* fix(otel): keep the served pricing basis on a cache hit's cost breakdown
A hit's breakdown now keeps service_tier, data_residency, vertex_location, and the discount and margin rates the served call recorded, with every amount at 0, so savings reports still read the basis. The hit header test now drives the live non-streaming header path.
* fix(otel): send zero breakdown headers on a cache hit and give hit spans a fresh call's cost keys
A hit's x-litellm-response-cost-original, -input, and -output headers now read 0.0 instead of going missing or carrying the full price. A hit's cost breakdown now comes from pricing the response as fresh, so every route exports the same keys, pricing basis, and rates with every amount at 0.
* test(integration): assert zero cost and the cache-hit flag on cache-served OTel V1 and V2 spans
* test: wait for the cache entry before the cache-hit request and build cost headers without a route_request double
* test: annotate the new cache-hit unit test locals with Final
---------
Co-authored-by: shrey kharbanda <shreshth@berri.ai>
Co-authored-by: jesus <jesus@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* fix(presidio): resolve overlapping entity spans in numbered tokens anonymization (#42130)
Mirror of BerriAI/litellm#42133 by @PINYOPATTANAWASANPORN, squashed onto main
with the unit test moved to tests/unit after the test tree relocation
* fix(presidio): label an overlapping run by its outermost detection, add integration coverage
A run of overlapping detections now takes its entity type from the highest-scoring
detection that no other detection in the run contains, matching Presidio's own
anonymizer, so a phone number with a date-shaped fragment inside stays PHONE_NUMBER.
The resolver returns typed detections instead of Any.
* fix(presidio): drop the redundant comment in the numbered token loop
* fix(presidio): bump a numbered token another part of the request already holds
Each message and content part numbers its tokens from 1 into one request-wide
pii_tokens map. Collapsing overlapping detections made an overlapping value's
token well formed, so it collided with the same type in another part and both
restored as the later value. A token that maps to a different value now moves
to the next free number, and a token for the same value is reused
* fix(presidio): give a clashing token a spare number instead of scanning for one
Scanning upward from the clashing number probed every token an earlier part
had claimed, so two parts with thousands of values of one type took seconds
of synchronous work. No token in the map is numbered above the map's size,
so numbers above both that size and the part's run count are always free
* test(presidio): make the anonymizer double reject detections the analyzer never returned
Pick the token in one conditional expression instead of reassigning it, and
have the scripted anonymizer answer 500 when the proxy forwards spans that
differ from the analyzer's, so the default-scope cells check what was sent
---------
Co-authored-by: PINYO PATTANAWASANPORN <pattanawasanporn@gmail.com>
Co-authored-by: yucheng-berri <293926326+yucheng-berri@users.noreply.github.com>
* fix(responses): run pre_call guardrails on every Responses WebSocket turn and bill each turn its own guardrail cost
* test(responses): keep the websocket unit test file on main's formatting
* fix(responses): type the websocket frame guard for the lint gates
* fix(responses): pass the websocket frame guard as a named parameter and mark the guardrail dict edits
* fix(responses): bill a guardrail-rejected Responses WebSocket frame as a failed request with its guardrail cost
* fix(responses): type the recorded guardrail records as Mapping for the lint gate
* fix(responses): bill a rejected websocket frame through the spend logger only so the connection keeps its rate-limit slot
* fix(responses): guard nested response.create frames, strip forged user_api_key_* frame metadata, copy guarded tools back and give each rejected frame its own spend row id
* fix(responses): mark the spend logger's bare-dict failure hook for the type gate
* fix(responses): reject response.create frames that carry input at both levels and keep billing of a rejected frame non-blocking
* fix(responses): name the non-blocking billing catch for the strict lint gate
* fix(responses): keep the admitted guardrail list on every websocket frame, a frame cannot replace it with its own guardrails list
* test(responses): wait for the key spend total before asserting it in the websocket guardrail cost test
* fix(responses): copy the admitted metadata shallowly per websocket frame, the OTEL parent span cannot be deep-copied
* fix(responses): reject a WebSocket turn a pre_call guardrail rerouted to another model
* fix(responses): bill a rerouted WS turn and carry Presidio pii_tokens from the frame guard to the socket unmasker
* fix(responses): default pii_tokens through default_factory, Python 3.13 rejects a mappingproxy dataclass default
* fix(responses): run the per-frame guardrail walk on managed Responses WebSockets too
* fix(responses): unmask managed Responses WebSocket replies with the frame guard's Presidio pii_tokens
* fix(responses): strip client guardrail cost records from WS frames so a blocked frame bills the real scan
* fix(responses): forward a WS frame without the fields a guardrail deleted
---------
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(anthropic): carry safeguard_results and stop_details in synthesized message_delta events
When the proxy renders a complete Anthropic Messages response as an SSE
stream for a client that asked to stream, the message_delta event carried
only stop_reason and stop_sequence. A real Anthropic stream puts
safeguard_results and stop_details inside message_delta.delta, so a client
such as Claude Code auto mode lost the server-side classifier verdict and
a refusal lost its stop_details. Both stream builders now copy those keys
from the source response through one shared helper when they are present.
* fix(guardrails): keep safeguard_results and stop_details when re-emitting a rewritten Anthropic stream
A post-call guardrail that rewrites an Anthropic stream (presidio, bedrock, tool permission, model armor) now carries the upstream message_delta safeguard_results and stop_details into the re-emitted stream. The router fallback message_start no longer leaks those final fields early
* fix(guardrails): keep stop_details and safeguard_results in rewritten Anthropic streams without explanations
A guardrail that rewrites a /v1/messages stream now carries stop_details (type and category) and the
dangerous_tool_use verdicts for the tool calls it still delivers, with every explanation dropped so a
masked stream never echoes the unmasked text through those fields.
* fix(guardrails): copy only upstream stop_details and forward-compatible safeguard_results into rewritten Anthropic streams
A rewritten stream now takes stop_details and safeguard_results from the
upstream message_delta alone, so a refusal the upstream sent without
stop_details is never given one by the OpenAI round trip. Unknown safeguard
types, statuses, reasons and extra fields pass through, a verdict follows a
tool call only when its input is unchanged, and tool_permission keeps the
explanations since it rewrites no text.
* fix(guardrails): apply the safeguard_results and stop_details rules to policy pipeline stream rewrites
* fix(router): deliver a refused stream built from a finished agentic turn instead of falling back
Websearch, MCP and advisor streams replay a complete reply whose tools already
ran, so a refusal fallback would run every tool again. Also give the recovered
response's message_start a null stop_details as the API sends it, drop
safeguard_results a masking rewrite cannot read, and add the audit cells for
synthesized streams.
* test(router): annotate the finished-turn refusal stream test locals with Final
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* feat(bedrock): apply bedrock_request_metadata_fields on the /bedrock pass-through routes
* fix(bedrock): move Invoke body requestMetadata into the signed header and match Bedrock's ASCII whitespace rule
* fix(bedrock): keep Invoke body requestMetadata in the body and answer bad metadata as AWS does
Invoke requests whose body already carries requestMetadata keep the resolved pairs in the body
instead of the 8,500-character signed header. Malformed caller metadata now answers 400 with the
ValidationException error type the AWS SDK expects, and drop_params drops bad pairs on both the
pass-through routes and /v1/chat/completions. The caller-pair dedupe is linear, and the chat
validator treats \s as ASCII whitespace, as Bedrock does
* fix(bedrock): match Bedrock's 400s and header limit on pass-through request metadata
Invoke metadata too long for the 8,500-character header now goes in the body,
Invoke requestMetadata null answers 400 like Bedrock, Converse type errors carry
SerializationException, violation messages name the failing key, and the reserved
user_api_key_ prefix is matched case-insensitively after trimming whitespace
* fix(bedrock): trim the Invoke metadata header to Bedrock's limit and drop bad requestMetadata under drop_params without the setting
* fix(bedrock): leave /bedrock pass-through bodies as sent when request metadata forwarding is off
Global drop_params no longer rewrites requestMetadata on the pass-through routes unless
bedrock_request_metadata_fields is set. Chat keeps dropping bad pairs under drop_params.
Adds a tests/integration wire test for the four pass-through actions and removes an unused
parameter from the Invoke header helper.
* test(bedrock): drive the pass-through metadata wire test with boto3's bearer key env var
* fix(bedrock): trim the pass-through invoke metadata header and stop Nova invoke from checking metadata it drops
* test(bedrock): cover pass-through metadata header trimming, malformed caller metadata, and chat drop_params on the wire
* fix(bedrock): match requestMetadata keys and values against the whole string so a trailing newline cannot pass the 256-character limit
* fix(bedrock): scan Claude invoke-with-response-stream passthrough frames with post_call guardrails (LIT-9618)
Invoke streams of Claude models now go through the same frame-by-frame relay as
converse-stream. Each chunk frame's Anthropic event is scanned by the /v1/messages
stream translation, a block ends the stream with an AWS exception frame, and a held
stream gets its masks written back into re-encoded chunk frames. Claude is told by
its model id or, behind an application inference profile, by the anthropic_version
its request body carries. Streams of other model families stay as they were
* test(bedrock): cover invoke stream vendor outage, upstream exception and release on scan (LIT-9618)
* fix(bedrock): hold Claude invoke streams for guardrails the relay cannot run and mark each scan as streaming (LIT-9618)
A guardrail released on scan or run in parallel left a Claude invoke-with-response-stream unscanned, while
converse-stream holds the whole stream and scans it once. Claude invoke streams now take the same buffered
path, with the rewrites written back into the chunk frames that carried each text.
Each invoke scan also sends stream true and the events relayed so far as responses, as /v1/messages does,
so a vendor can tie the scans of one streamed reply together.
* refactor(bedrock): mark the request dicts the buffered stream paths hand to post_call hooks (LIT-9618)
* fix(bedrock): scan a Claude invoke stream the upstream breaks off as a finished stream (LIT-9618)
An invoke stream that ended on an upstream exception, without a message_delta stop_reason, was scanned
as still in flight, so its tool input never reached the guardrail and a held stream released it unscanned.
The scan and its scan key now close such a stream the way converse-stream treats the end of its upstream.
* test(bedrock): pass the converse-stream endpoint where the tool call without input test frames its stream (LIT-9618)
* test(bedrock): expect the ValidationException main now returns when a guardrail blocks a Claude invoke stream before its first frame (LIT-9618)
* test(bedrock): type the Claude invoke event helpers in the passthrough guardrail tests
* test(bedrock): type the remaining parameters of the Claude invoke stream tests
* refactor(bedrock): bind the buffered invoke rewrite in its condition
---------
Co-authored-by: yucheng-berri <293926326+yucheng-berri@users.noreply.github.com>
* test(integration): Converse stream post_call guardrail relays frames as they arrive (LIT-9515)
* fix(bedrock): stream Converse passthrough frames through post_call guardrails instead of buffering (LIT-9515)
A post_call guardrail that scans through apply_guardrail now runs on
/bedrock/.../converse-stream the way it runs on /v1/messages: frames reach the
caller as the upstream sends them, every fifth frame and the end of the stream
are scanned, and a block after the first frame ends the stream with an AWS
exception frame. streaming_buffer_until_moderated still holds the stream and
applies masks per content block. Guardrails that need the whole response
(own async_post_call_success_hook, mask_response_content, native hooks, policy
pipelines) keep the buffered path.
* fix(bedrock): apply held Converse stream masks when the upstream ends with an exception frame (LIT-9515)
* fix(proxy): log a relayed passthrough stream's spend before its mid-stream failure hook (LIT-9515)
* fix(bedrock): close and report every way a relayed Converse stream ends (LIT-9515)
- The framer ends the stream with an empty item once the upstream closes, so held masks are applied whether the
upstream ends with messageStop, an exception or error frame, or a dropped connection
- A guardrail HTTPException after the first frame reaches the relay, which logs the stream's spend, runs the
failure hook and ends the stream with the exception frame, instead of ending inside the guardrail with no
failure row
- A failure before the first frame waits for the stream's spend to be logged, as a mid-stream failure does
- Closing the response before its body starts closes the guardrail layers and the upstream
- Each stacked guardrail layer gets its own copy of every frame, so an earlier layer's end-of-stream rewrite never
reaches frames a later layer still holds
- Deltas whose text ends in half a surrogate pair are decoded with the standard JSON parser
* fix(bedrock): relay a Converse frame whose payload is not UTF-8 instead of ending the stream (LIT-9515)
* test(integration): cover Converse stream relay failure, end-of-stream-only, stacked, disconnect and burst cells (LIT-9515)
* fix(bedrock): estimate prompt tokens for a passthrough stream cut before Bedrock reports usage (LIT-9515)
Bedrock sends usage only in the last frame of a stream, so a stream ended by a guardrail block or a client
disconnect logged 0 prompt tokens while AWS still charges for the prompt. The stream collector now estimates them
from the request the way chat completions streams do
* fix(bedrock): read tool result blocks without recursion in the passthrough prompt estimate (LIT-9515)
* fix(passthrough): keep the spend flush running when the wait for it is cancelled (LIT-9515)
* fix(bedrock): keep a converse-stream buffered when its guardrail releases a held stream on scan (LIT-9515)
A held post_call guardrail with streaming_buffer_release_on_scan releases each
window as received once its scan passes, so the relay delivered the original
text instead of the guardrail's masks. Such a stream keeps the buffered path,
which applies masks per content block as on main.
* fix(bedrock): keep a converse-stream buffered when a post_call guardrail runs in parallel (LIT-9515)
The relay ran every guardrail in config order, so a run_in_parallel guardrail
listed before a masking guardrail scanned the unmasked text. Main's buffered
path runs parallel guardrails after the sequential ones and discards their
rewrites, which the relay cannot reproduce, so these streams stay buffered
* fix(bedrock): mark Noma v2 relay scans like chat and stop reading Bedrock when a held stream's client leaves
- Noma v2 scans on the converse-stream relay carry stream, and the last one
carries the finished Converse response with its stopReason, as chat sends
them, so Noma keeps one record per reply. Other guardrails get the request
unchanged. When the last sampled scan already covered the whole answer, the
end-of-stream scan still runs for Noma so it gets the finished answer
- A held relay stream whose client leaves before the first frame stops
reading Bedrock and returns 499, like chat. When Bedrock sent nothing yet,
the budget reservation and the parallel-request slot are released
- A cancelled wait for the first streamed chunk now stops the upstream read
instead of leaving it running
- A Noma v2 scan cancelled before Noma answers records not_run instead of
success with an empty response
* fix(bedrock): send Noma the finished Converse reply's tool calls and reasoning as Converse blocks
* fix(bedrock): give Noma the ended scan of a Converse reply that only calls a tool without input
* test(bedrock): expect the ValidationException main now returns for a held block before the first frame
Main's AWS-shaped passthrough errors (#46522) answer a guardrail block before the first frame with x-amzn-errortype ValidationException, so boto3 reads the code and the guardrail message instead of a bare 400.
* fix(bedrock): keep held masks when an earlier sampled guardrail ends the Converse stream with its message
A sampled guardrail that blocks with disable_exception_on_block ends the stream with its message as raw
event-stream bytes. A held guardrail after it on the same key only read frame objects, so it never saw the
messageStop, skipped its write-back and released the held text without its masks. The Converse handler now
reads those bytes back into frames, so the held guardrail sees the end of the message and masks the text
* fix(proxy): free the parallel slot when a caller leaves a Converse relay holding part of the first frame
The relay skipped releasing the slot and budget reservation whenever a spend
flush was scheduled, but a flush of a partial first frame builds no response
and logs nothing, so the slot stayed held until it expired. The flush now says
whether it logged the stream.
---------
Co-authored-by: yucheng-berri <293926326+yucheng-berri@users.noreply.github.com>
* test(e2e): event_loop_stalls perf case reads the gateway's asyncio debug log and fails on a loop hold over 100 ms that is not on its allowlist
The gateway runs the stdlib loop in debug mode (new LITELLM_EVENT_LOOP=asyncio knob next to
PYTHONASYNCIODEBUG=1), the case replays the mixed body mix at 1 rps and groups asyncio's
'Executing <Task ...> took X seconds' lines by coroutine and location. Known holds are listed with
the count they show today; anything else, or a known one past its count, fails the case. Opt-in
through E2E_PERF_EVENT_LOOP=1 so the nightly build skips it
* test(e2e): event_loop_stalls reads the pod list and pod logs through e2e_http instead of a raw httpx client
* test(e2e): event_loop_stalls charges a hold to garbage collection when the gateway's debug-mode gc log accounts for it
The proxy registers a gc.callbacks hook only when the event loop runs in debug mode and logs every
collection over the loop's slow-callback threshold (generation, collected count, seconds). The case
parser pairs such a pass with the hold line that follows it and reports the hold under a bounded
garbage collection bucket instead of the task asyncio happened to blame, which made 7 of 10 local
runs fail on a different task name each time. Holds the collector does not account for keep the
strict allowlist, which gains the prisma first-query warm-up and the second pub/sub subscriber
* test(e2e): event_loop_stalls allowlist caps set at twice the ten-run maximum
---------
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* perf(cli): start lite without initializing the litellm SDK
lite and litellm-proxy now enter through a small litellm_cli package that imports the CLI against an uninitialized litellm package and runs the SDK __init__ in place only when something first uses it, so commands start in a fraction of a second instead of loading about 3,000 modules first.
* perf(cli): import only the lite subcommand being run
Each lite subcommand now loads its module when Click looks it up, so
lite logout no longer imports the autoroute wizard, agents, or the other
command groups. importlib.import_module and attribute reads on the SDK
also initialize it on first use, and an already imported litellm is
kept as is.
* fix(cli): initialize the SDK on a star import and test command bodies in thin mode
A from litellm import * after lite loaded the CLI used to see only the names the thin package had, so it now runs the full SDK init first. The thin-vs-full parity test now also runs logout and whoami, not just --help
* fix(cli): keep the import hooks until the lazy SDK init finishes
A thread that imported SDK code after the init restored the interpreter
imports could hold that module's import lock while waiting for the init,
and the init then waited on the same lock, so both threads hung. Other
threads now wait for the init before they load any new SDK code
* fix(cli): retry a failed lazy SDK init on the next use instead of leaving it half-run
* test(cli): bind the patched credentials client and autoroute dir in an order-independent way
* fix(proxy): unset DB environment variables removed from the row on every worker by row revision
* fix(proxy): order the environment variables revision migration after main's and set its lock timeout
* fix(proxy): snapshot the environment at the first DB sync even when no environment_variables row exists yet
* fix(proxy): close gaps in DB environment variable reconciliation
Grant the revision sequence to every role so a writer role with only
table grants can still save environment_variables. Unset SSO keys that
only the environment_variables row stored on the next SSO sync, while
keeping keys the SSO settings table holds. Skip a revision-less row once
a revision has been applied, so a stale cached row cannot bring a
removed key back. Keep plaintext and undecryptable values when rotating
the master key instead of dropping them from the row.
* fix(proxy): keep env var revisions monotonic across DB copies and restores
Stamp at least the old revision plus one on update so a copied row keeps
rising, re-read a stale cached row from the writer and apply it when the
database itself went back, unset dropped SSO keys when sso_settings are not
loaded from the DB, and keep env values master key rotation cannot decrypt
exactly as stored
* fix(proxy): skip an env var row read before an overlapping sync that already applied a newer one
* test(proxy): cover a read replica back on an older row and an SSO client removed from DB environment_variables
* fix(proxy): keep config reads up when the writer is down, find the revision sequence by table schema, read the writer before rewriting env vars, and add the trigger under db push
* test(proxy): model the writer database in the retry policy config update fake
* fix(proxy): recheck the environment_variables row revision right before applying it
A sync that started later could read a stale environment_variables row
before an earlier sync applied a newer one, then pass the read-order check
and apply the stale row, bringing back keys the newer row removed. The
behind-the-applied-revision check now runs at apply time, re-reading the
writer when the row is behind, and a writer row read while another sync
applied a newer revision is dropped.
The config sync tests now inject a fake database client and use the real
get_config_param and its cache instead of replacing get_config_param.
* test(proxy): keep encryption real in the SSO replica-lag test
* test(proxy): inject a fake Prisma client in the env var replica-lag test
* fix(proxy): keep the applied environment variables when the writer database is out of connections
* test(proxy): cover a writer database out of connections in the env var integration tests
A turn whose admission check raises an unexpected error still gets a 500 error frame, but the socket
now stays open, like a refused turn. Closing it cut any turn still streaming on the same socket, which
a client that sends its next response.create early would lose.
A refused turn's failure row now takes the router's tags only when every tag is a string, so the
fallback never copies a caller's unvalidated tags onto the row.
* fix(openai_like): surface provider response headers on non-streaming chat completions
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(openai_like): mark the header test locals Final
* fix(proxy): leave out response header values HTTP cannot carry
Non-streaming OpenAI-like chat now forwards every provider header, so on the
httpx transport a provider header value outside latin-1 made the proxy answer
500 after the provider had served the call. get_custom_headers now leaves such
a value out and keeps the other headers, which also ends the same 500 on
streaming
Adds integration cells for the provider request id on non-streaming chat and
the Responses bridge, LiteLLM headers an upstream cannot overwrite, spend
parity under an upstream cost header, the latin-1 case, and the logging
callback payload
* fix(cost): ignore an upstream response cost header that is negative or not a number
* test(integration): cover invalid upstream cost headers and non-latin-1 headers on every endpoint
* fix(prometheus): keep deployment metrics when a provider rate-limit header is not a number
* test(prometheus): type the rate-limit header test and drop its clock reads
* test(integration): cover deployment metrics under an unparseable provider rate-limit header
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* feat(anthropic): workload identity federation and pluggable identity sources
Backend half of #38818 (internal copy of the fork PR #38013), rebuilt as one
commit on top of litellm_internal_staging without the dashboard changes.
Deployments on anthropic/ without a static api_key can exchange an OIDC
workload assertion for a short-lived sk-ant-oat01 token through a shared
RFC 7523 JWT-bearer engine. The assertion comes from a mounted token file,
an env token, a LiteLLM-signed issuer, or Keycloak, chosen per deployment,
per named credential, or through ANTHROPIC_IDENTITY_SOURCE. The federation
fields are server-owned: refused inline in request bodies and on
POST /model/new, proxy-admin only on credentials, and the token exchange
is pinned to api.anthropic.com unless LITELLM_ANTHROPIC_WIF_ALLOWED_HOSTS
adds a host. GET /credentials/{name}/jwks exports the public key set of a
LiteLLM-signed credential for the Claude Console.
The OpenAI federation trio from #39613 rides along on the backend side with
the same server-owned handling.
Fixes#28607
Resolves LIT-6107
Co-authored-by: derhornspieler <15236687+derhornspieler@users.noreply.github.com>
* fix(anthropic): let batch-result downloads mint from deployment params and accept host:port allowlist entries
The files handler enabled workload identity on batch-result downloads but never received the
deployment's litellm_params, so a deployment authenticating through a named credential could only
mint from process-wide env vars. It now threads litellm_params through to the auth header the way
the batch retrieve path already does.
LITELLM_ANTHROPIC_WIF_ALLOWED_HOSTS entries written as host:port were read by urlsplit as a scheme,
so the allowlist kept the raw entry while the exchange compared bare hostnames and refused the
gateway. Entries are now parsed as network locations whether or not they carry a scheme.
* fix(types): move the WIF kwargs key sets to a leaf module so the kwargs funnel imports without a cycle
* test(anthropic): pin case-insensitive matching of WIF exchange-host allowlist entries
* fix(anthropic): end workload identity federation errors without a period so the router suffix reads cleanly
* fix(proxy): decrypt stored litellm_params before the WIF write gate
* fix(proxy): hide WIF secret references from /health output
* fix(proxy): keep the proxy error shape on credential endpoint refusals
* fix(proxy): hide identity token file paths from /health output
* feat(openai): honor per-deployment workload identity on non-chat routes
Image generation, audio speech and transcription, moderation, text
completions, files, batches, and fine-tuning built their OpenAI client from
the env vars alone, so a keyless deployment carrying
openai_identity_provider_id, openai_service_account_id, and
openai_identity_token_file failed with the api_key error on every route
other than chat, Responses, and embeddings. Those handlers now build their
client through the shared workload identity resolver, and the managed batch
file-access credential extraction keeps the trio
* fix(openai): cache workload identity clients and let deployment identity outrank process-wide keys
Non-chat handlers built a fresh SDK client per call, so files, batches,
fine-tuning, text completion and sync moderation exchanged the OIDC token on
every request. The builders now reuse the cached workload identity client per
identity, base, timeout, retries and organization; static-key clients are
still built fresh.
A process-wide OPENAI_API_KEY (or litellm.api_key / litellm.openai_key) no
longer silently overrides a deployment's openai_identity_* params. An explicit
distinct api_key on the deployment or call still wins.
* fix(openai): keep shared workload identity clients on the process http session
The workload identity client is cached and shared across requests, so it must
never borrow the transport a single caller passed in. Static-key clients keep
using the caller's http client; federated clients always ride the process-wide
session. Also makes the http client helpers on BaseOpenAILLM public since the
builder now needs them from outside the class.
* test(openai): pin partial deployment identity beating the process key
* test(openai): type the managed-object invoke callback
* fix(anthropic): rename the federation workspace param so Bedrock's anthropic_workspace_id keeps working
The Bedrock Claude Platform route already reads anthropic_workspace_id from
optional_params, so banning that spelling as a server-owned federation
parameter broke a pre-existing client capability. The federation field is now
anthropic_federation_workspace_id (env ANTHROPIC_FEDERATION_WORKSPACE_ID),
which restores the base branch's behavior for Bedrock callers, drops the
Bedrock-specific hint from the refusal message, and deletes the unconditional
ban constant that no longer had a reader
* fix(auth): share one exchanged token across workers reading the same assertion
Anthropic accepts each identity assertion exactly once, so two uvicorn
workers reading the same token file both minting from it means the second
exchange is denied with jti_reused. Minted tokens now land in a per-user
0700 cache directory guarded by a file lock, so workers on the same host
reuse one exchange until the token expires or the assertion rotates. A 401
is only retried when the re-read assertion actually differs, and the denial
hint explains jti_reused. LITELLM_TOKEN_EXCHANGE_CACHE_DIR moves the cache
and an empty value disables it
* feat(openai): honor deployment workload identity on image edits, vector stores, containers, videos, and evals
The httpx-based OpenAI handlers built their Authorization header from the
deployment api_key or OPENAI_API_KEY alone, so a keyless deployment carrying
openai_identity_provider_id, openai_service_account_id, and
openai_identity_token_file sent "Bearer None". They now resolve the bearer
through resolve_openai_bearer_token, which exchanges the deployment's
identity token the same way the SDK-based routes do.
A deployment that carries only part of the identity trio now raises a 500
naming the missing field instead of silently falling back to the process
key, and the SDK client builders take a factory for the static-key http
client so the federated branch never touches litellm.client_session.
* fix: keep anthropic federation from being shadowed or leaked
An empty or whitespace-only ANTHROPIC_API_KEY counted as set, so a federated
deployment sent an empty x-api-key on every call instead of minting a token.
Blank values now read as unset, and a real static key on a federated deployment
logs once that it outranks federation and nothing is being federated.
The exchange-host allowlist matched hostnames only, so a second process on
another port of an allowed host was trusted with the workload's identity token.
An entry that names a port now trusts that port alone, while a bare host still
trusts every port.
The shared token store exists so the workers reading one projected token file do
not each spend its single-use jti. A source that mints its own assertion per
exchange shares nothing with another worker, so it no longer writes a live token
to disk for a lookup that can never hit.
* feat(openai): resolve per-deployment workload identity for the assistants routes
OpenAIAssistantsAPI built its OpenAI and AsyncOpenAI clients from a locals() scrape, so a deployment carrying openai_identity_provider_id, openai_service_account_id and openai_identity_token_file never exchanged a token. Thread litellm_params through the class and build both clients with the shared workload identity builders.
* fix: unlink a staged token file a failed write leaves behind
The 401 denial hint now also says federation ignores ANTHROPIC_WORKSPACE_ID, which the Bedrock Claude platform provider already reads.
* refactor: move anthropic jwks derivation behind a provider-owned tagged union
* feat(openai): resolve per-deployment workload identity for image variations
Build the image variations client through the shared workload identity
builders instead of scraping locals(), and give the async path the async
http session it was missing. Let a federation trio serve the model-less
batch fallback, and hold the standard LRU size of exchanged tokens.
* fix: unlink the staged token file when its write fails at close
A buffered write only reaches the disk when the handle closes, so a full disk surfaces at close and left the staging file behind holding a usable token.
* fix(anthropic): close the staging descriptor before writing the shared token file
* fix(wif): judge federation writes by what they set, not what is stored
The admin gate read the stored deployment, so a team admin lost edit, delete
and Test Connection on any deployment carrying federation params. It now
returns early unless the submitted fields touch the federation surface, and a
Test Connection probe that points the deployment at its own api_base is still
refused, with the 403 no longer wrapped into a 500
The rest of the same review pass: POST /model/new refuses only a blocking
value of `blocked`, so a client that always sends `blocked: false` is not
turned away; a request body can no longer pick which federated identity to
mint as by naming a stored credential; an advisory refresh the executor
refuses disarms the entry instead of wedging the identity until the follower
timeout; the static-key shadow warning resolves its env fallback inside the
cache instead of once per request; credential writes drop nulls before
storing them; the token exchange validates the endpoint URL before reading an
assertion and keeps refusing redirects across a client heal; /health hides
every server-owned federation field from non-admins; and the async create_file
and create_batch paths say which setting is missing when the provider resolves
no URL
* style(tests): keep two blank lines before the batch_cost_is_final block
* test(batches): type the workload identity fallback test fixtures
* fix(moderation): stop validating unrelated kwargs on the sync workload identity path
* fix(auth): refuse server-owned workload identity keys in array form and inside a client router config
* fix(openai): keep a deployment's own key ahead of workload identity on every route
Thread the caller's own api_key into the identity resolver for embedding, transcription, moderation,
image generation, and assistants so a deployment key equal to the process key still wins. A partial
identity trio now falls back to a configured static key, unset max_retries keeps the SDK default, and
/v1/decisions plus OpenAI count_tokens resolve the exchanged bearer for identity-only deployments
* fix(containers): pass deployment params to container file auth headers
Container file routes built their auth headers without the deployment's
litellm_params, so a keyless deployment with an OpenAI workload identity
sent Bearer None on upload, list, retrieve, delete and content calls
* fix(openai): keep process-wide key precedence and move token exchange off the event loop
Restore the merge base order where an explicit or process-wide OpenAI key wins over
federation, so exporting OPENAI_API_KEY keeps working the way it did before.
Run the header-building validate_environment calls (image edit, videos, containers,
vector stores, vector store files) and the decisions and System One request
preparation in a worker thread, so a slow token exchange no longer stalls the loop.
Federate realtime websocket sessions, live sessions, realtime client secrets and
transcription sessions, and the OpenAI moderation guardrail when no key is set.
* fix(openai): answer 401 for a partial identity trio and federate realtime health checks
A deployment with part of the workload identity trio and no static key now
fails as an auth error naming the missing params, so the router does not retry
it as a server error. The realtime health check sends the exchanged bearer for
a keyless federated OpenAI deployment, and API token counting reads the same
static keys as every other route before minting.
* fix(openai): keep local token counting without an identity trio and federate the moderation guardrail from its own config
* fix(openai): skip a partial-trio moderation guardrail and let vault keys win in token counting
* test(integration): refuse caller-chosen identity params in user_config and bracketed form keys
* fix(openai): send the honored secret manager key from SDK client builders
A deployment whose OPENAI_API_KEY lives only in the secret manager now sends that key
on image, assistants, files, batches, and other SDK-built routes instead of failing with
no api_key, matching what the workload identity resolver already treats as a static key
* test(openai): check the token exchange runs off the event loop instead of timing the loop
---------
Co-authored-by: derhornspieler <15236687+derhornspieler@users.noreply.github.com>
* fix(azure): add responses_api_version so the Responses route stops using the deployment chat api_version
* fix(azure_ai): fold Grok reasoning tokens into completion_tokens like the xai provider
* fix(utils): refresh provider model sets after register_model
* fix(cost-map): align azure gpt-6.1-sol reasoning effort flags with the openai row
* test(e2e): cover Azure Responses on a deployment pinned to a dated chat api_version
* fix(utils): keep router deployment ids out of provider model sets
register_model now refreshes the provider sets, so the per-request and
router registrations keyed by a deployment id put those opaque ids into
wildcard model listings. Those callers now pass
add_to_provider_model_sets=False.
* fix(azure): carry responses_api_version through the chat to Responses bridge
completion() dropped responses_api_version from the litellm_params it hands
the Responses bridge, so azure/responses/<model> on a dated chat api_version
still built the 404ing /openai/responses URL. Also cover the auto-router
marker never forwarding the new param.
* fix(azure): stop listing a wrong set of reasoning efforts when none is rejected
The message said only low, medium and high are supported, which is wrong for
models that also take xhigh such as gpt-6.1-sol.
* test(azure): use the latest grok and gpt-5.4 rows in the new azure tests
* fix(proxy): list each wildcard-expanded model once
A provider set can hold a model both bare and provider-prefixed, which now
happens after any priced request registers its openai/<model> key, so the
openai/* listing showed that model twice.
* fix(azure): route Responses past dated GA api_versions by default
A deployment pinned to a dated GA api_version such as 2024-10-21 for chat now sends Responses calls (create, get, delete, cancel, input_items and the azure/responses/ bridge) to the default Responses version, because those dated versions have no Responses route on Azure and 404. responses_api_version still overrides it, -preview, v1, latest and preview versions are kept, an api-version in the api_base query still wins, and chat keeps api_version as before
* chore(azure_ai): mark the grok chunk_parser dict as an override contract
* fix(azure): rewrite only the route's /openai segment when moving a call to v1
The v1 rewrite in the Azure URL builder replaced every "/openai" substring in the path, so a gateway
api_base like https://apim.example.com/openai-gw/team turned into /openai/v1-gw/team/openai/v1/responses.
Dated GA Responses calls now land on the v1 route by default, which made this reachable for them too,
so the rewrite now only touches the last /openai path segment.
The Responses version resolver no longer duplicates the builder's default for an unset api_version.
Tests cover the gateway api_base case, the env AZURE_API_BASE query-string precedence, and the e2e suite
now has one test for the dated GA default and one where only responses_api_version can make the call work
* test(integration): cover Azure Responses api_version, Grok reasoning spend, wildcard listing and reasoning_effort none
Wire tests pin the Azure route and query for chat, Responses and the azure/responses bridge on a dated GA api_version, and check responses_api_version never reaches the request body. Cost cases check azure_ai Grok folds reasoning tokens into completion_tokens and spend, streaming and not. A config-driven proxy checks /v1/models next to an openai/* wildcard lists each id once and never a deployment id. Azure gpt-6.1-sol rejects reasoning_effort none before reaching upstream and drops it with drop_params
* fix(utils): list ai21 chat and vertex code models under their provider wildcard
register_model now files ai21 chat models into ai21_chat_models, which models_by_provider never read, so a registered ai21 chat model dropped out of ai21/* listings. models_by_provider now reads ai21_chat_models and the vertex code chat and code text sets too.
* fix(main): keep caller-chosen model names out of provider wildcard listings
---------
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
* fix(sso): concurrent first sign-ins land on the account the winner created, worker-kill test pins half its streams to each worker
* test(integration): a worker kill returns once the worker is dead, the reap waits at block exit
uvicorn can take a full worker healthcheck to reap a worker that died mid-ping, so kill() held
the caller that long and peers held on the surviving worker hit their client timeout
* fix(sso): a sign-in that loses the first-insert race joins the default teams
The losing sign-in could read the winner's row before new_user added the default teams, so its
CLI key was minted with no team. The race tests now run the real new_user against a fake user table
* fix(proxy): answer 400 for malformed tools on count_tokens instead of crashing
A non-list tools value crashed the vector store lookup during auth and the
built-in tool cost lookup in the failure logging path, so the proxy answered
an unhandled 500 and closed the keep-alive connection. count_tokens now
answers a 400 invalid_request_error for a malformed body, drops malformed
tools when drop_params is on, and both lookups skip values of the wrong shape.
* fix(proxy): skip non-list tools in the tool allowlist check
Also validate the count_tokens body without recursion, which the recursion detector rejects.
* test(integration): restore the int tools case on count_tokens
The proxy now answers a non-list tools value with a 400, so the case is deterministic again.
* fix(proxy): reject malformed tools at the LLM route entry and keep vector store checks on malformed ids
* fix(proxy): apply drop_params like core when dropping malformed tools, and reject malformed tool vector store ids
Check tools after request setup so the Claude Code default and the logged call type apply, read the request,
global, and model group drop_params the way core does, drop tool_choice and friends with tools, cover Gemini
generateContent, and answer 400 for a malformed tools[].vector_store_ids with the strict flag off
* fix(proxy): keep the request's call type when logging a gate rejection after request setup
The body-shape fallback relabeled a /v1/messages rejection as acompletion and a list-input /v1/responses
rejection as aembedding, so it now only applies when the logging object carries no known call type
* fix(proxy): share the malformed tools check across tool routes and honor drop_params per entry
* fix(proxy): accept null vector_store_ids, read every named store id in auth, check the responses websocket
* refactor(proxy): leave the responses websocket out of the request-level tool check
* fix(proxy): let a request's additional_drop_params replace the deployment list and keep compact's call type on rejections
* fix(proxy): read router default drop lists and deployment ids in the malformed tools check
* fix(proxy): leave malformed tools to the router when it can pick another deployment
Fallbacks, auto-routers, model strings, comma batches and team model names
let the router pick a deployment after the proxy checks the request, so the
proxy now forwards a malformed tools field there as main does, and only
rejects it when the deployment it checks is the one that serves. Token
counting never reroutes, so count_tokens and input_tokens keep reading the
group they count with. additional_drop_params names are read the way core
reads them, so a list with a non-string entry still names its strings.
* test(integration): cover malformed tools on reroutable models, router defaults and deployment ids
* fix(proxy): resolve malformed tools at the deployment the router picks
* fix(proxy): keep provider-repaired tool schemas, honor fallback_on_unsupported_params, skip cooldown on malformed tools
Empty schemas and JSON-string schemas pass the tools check the way litellm's provider paths already read them. The rejection is a BadRequestError subclass carrying the deployment's provider, so fallback_on_unsupported_params: false keeps the 400, and the router never cools a deployment down for a request litellm rejected before calling it. Adds the audit's pre_call guardrail and cross-provider bridge cells.
* test(integration): expect a 400 for an empty-string tools field on /v1/messages
Real Anthropic rejects tools "" with a 400 on main too, so the case joins the malformed tools test
* fix(router): keep web search routing for tools passed as a tuple
* fix(router): keep a deployment's own tools when drop_params drops a malformed request tools
* test(router): list _merged_tools as covered through the deployment tools merge tests
* fix(router): drop the request's tool settings with its malformed tools when deployment tools replace them
* fix(router): skip a tool whose type is not a string in the web search filter instead of raising TypeError
* fix(prometheus): keep deployment health metrics clean on a malformed tools rejection
* test: type the new malformed tools tests
* test: mark new test locals Final and type kwargs dicts
* fix(guardrails): accept Gemini tools and Anthropic content blocks in generic_guardrail_api, name the error type on fail-open
GuardrailToolParam required a type key, so Gemini native tools
(function_declarations, googleSearch) failed validation inside the hook
and either returned a 500 or skipped the scan when fail_on_error was
false. The type key is now optional and left out of the forwarded tool
when absent, so an echoing guardrail server cannot write type: null into
the provider request.
structured_messages was re-validated against the chat completion message
types, which silently emptied tool_result, tool_use, image and document
content blocks to content: [] and dropped undeclared keys before the
guardrail saw them. The request model now forwards rows as opaque
mappings, which also retires the write-back shim that restored the
original rows for echoed-back entries.
The fail-open critical log line now carries error_type so a timeout, a
refused connection and an upstream HTTP status can be told apart without
the traceback.
* fix(httpx): retry aiohttp server disconnects the way httpx does
aiohttp raises ServerDisconnectedError when a kept-alive connection is
closed by the peer before a response arrives, which is what a load
balancer idle timeout shorter than AIOHTTP_KEEPALIVE_TIMEOUT produces.
The transport mapped it to httpx.ReadError, which the handler's
connection-error retry does not cover, so the request failed instead of
being resent on a fresh connection. It now maps to
httpx.RemoteProtocolError, the error httpx itself raises for the same
condition, so the existing retry fires.
* test(httpx): build the stub server reply in one bytes literal
* fix(httpx): resend a dropped connection's request as sent and keep a streamed resend readable
The connection-error retry closed its fresh client before a streamed body was read,
so an aiohttp disconnect now billed a second upstream request and still failed. It
also skipped masking on a status error from the resend, resent PATCH and DELETE as
POST, dropped multipart files and ignored the handler's ssl_verify.
The retry now resends the request the first attempt built, on a client with the
handler's TLS settings, closes that client when a streamed body is closed, and masks
a status error the same way the first attempt does.
* fix(guardrails): log HTTPStatusError, not the masking wrapper, as the fail-open error_type
* fix(guardrails): resend a generic guardrail check cut before any response
A load balancer that closes an idle keepalive connection made the guardrail POST
fail with "Server disconnected", because #46336 only resends POST on ConnectError.
AsyncHTTPHandler.post now takes a keyword-only idempotent flag, defaulting to
False, that gives a RemoteProtocolError the same single fresh-client resend PUT
and DELETE get. Only GenericGuardrailAPI.apply_guardrail opts in, since a
guardrail check is a pure verdict over the request. Every other POST keeps the
#46336 rule
The fail-open log line now puts error_type at the end so the pre-PR text stays
contiguous for existing alert matchers
test_rows_reach_the_guardrail_verbatim now reads the body off a real local
guardrail server instead of mock call args
* fix(http): keep a mid-body server disconnect a ReadError and move socket tests to integration
A ServerDisconnectedError raised while reading a response body now stays
httpx.ReadError, as it was before this PR. Only a disconnect before any
response maps to RemoteProtocolError, which is the case the resend rule is
about. Stream readers that end on ReadError, like the SAP stream iterator,
keep their old behavior
The loopback-server disconnect, TLS and guardrail resend tests move to
tests/integration/sdk/test_disconnect_resend_wire.py, since tests/unit is
in-process only. Unit keeps in-process MockTransport checks for the idempotent
POST resend, masking on the resend, streamed and buffered client closing, the
guardrail resend, and the guardrail receiving rows verbatim
* fix(guardrails): resend a reset guardrail check and keep typed echoes on texts
An idempotent POST now resends once on any network error after connect
(ReadError, WriteError, RemoteProtocolError), so a guardrail behind a load
balancer that answers a stale keepalive with a TCP reset rides it out too.
That makes the aiohttp ServerDisconnectedError remap unnecessary, so it goes
back to ReadError. This stops PUT and DELETE from stacking litellm's resend on
top of aiohttp's own idempotent retry (4 sends instead of 2) and stops Vigil
Guard from replaying its analyze POST on a disconnect
Every resend now runs inside the method's own error mapping, so a timeout on
the resend is a litellm.Timeout like the first attempt and a status error is
masked once by the caller
A guardrail server that parses rows with its own message types and echoes them
back without the keys those types lack (annotations, tool_calls[].index) is
read as an echo again, so its texts redaction reaches the LLM. A row only counts
as an echo when no value changed and every key the chat completion types
declare is kept
The wire tests for the PUT/DELETE resend, its masking and its TLS settings now
run on the httpx transport, where litellm's resend is the only retry, and the
POST and guardrail tests cover both transports and both cut styles
* test(guardrails): black box wire tests for generic guardrail shapes, fail-open error type and idle cut resend
Integration tests against an owned proxy and scripted peers. Gemini native tools reach the guardrail and Gemini as sent, the guardrail sees chat rows and an Anthropic image block as sent, an echoing guardrail leaves a /v1/messages request unchanged, the fail-open log line names ConnectError, Timeout or HTTPStatusError, and a guardrail check cut on an idle pooled connection is resent once on each worker. Each fails at the merge base and passes at the head
* fix(guardrails): check echoed rows without recursion and keep pytest.raises to one statement
* refactor(guardrails): move the dropped-connection guardrail resend out of this PR
The resend on a dropped or hung generic guardrail call, and the shared AsyncHTTPHandler retry rework behind it, move to a follow-up PR that adds a per-guardrail num_retries setting. This PR keeps the request and response shape fixes and the error_type on the fail-open log line
* test(guardrails): pick the closed guardrail port from the integration port allocator
Reading a port back from a port-0 bind trips TQ012, so the fail-open connect error case now takes its unused port from allocate_port
* feat(guardrails): forward structured_messages as sent only behind forward_messages_as_sent
generic_guardrail_api keeps main's OpenAI-format rows and echo comparison by default. Setting forward_messages_as_sent: true on the guardrail sends rows as the client sent them and treats an echo that only drops undeclared keys as unchanged
---------
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(mcp): stop proxy-started legacy listeners from consuming server rpm
The listener the proxy opens on notifications/initialized for legacy MCP
sessions ran the server rate-limit admission once per target, so every
session spent an rpm slot before the client sent anything. Admission is
now injected into the listener and applied only to client-initiated
subscriptions (subscriptions/listen and resources/subscribe). Authorization
and its periodic recheck still run on every path.
* test(mcp): key the fake admission window by caller and server
* fix(mcp): stop subscription catalog refreshes from consuming server rpm
A subscription listener lists the caller's catalog to diff list_changed
events, at setup and on every upstream event. Those listings went through
the normal per-server admission, so against an upstream that declares
listChanged each legacy session spent several rpm slots before the client
asked for anything. The listener now marks its targets as already admitted
for its own lifetime, so its refreshes reuse the subscription's admission.
Client requests on the session are still metered.
* test(mcp): exercise subscription admission through the proxy rate limiter
* fix(proxy): keep a retried /v1/messages stream's spend log by logging the superseded attempt as a failure
A /v1/messages stream the router retries before any content shares one logging object across both attempts. The superseded attempt's end-of-stream logging ran as a success (a usage-only or placeholder result) and claimed the request's one-shot success log, so the recovered stream's spend row was either missing or written under the superseded attempt's message id. The passthrough logging path now logs an Anthropic stream whose frames carry an `event: error` as a failure with whatever usage the provider reported, the way a dropped stream already does, and the router closes the superseded stream before it opens the retry so that logging is queued first.
* fix(proxy): log a provider error frame through an attempt-scoped logging copy
The error-frame failure stashed the superseded attempt's usage on the logging
object the retry shares, so the retry's success log carried the failed
attempt's token counts. The failure now logs through a copy of the logging
object with its own model_call_details, and the helper moves into the
Anthropic passthrough logging handler so the proxy streaming handler only
routes by endpoint type.
* fix(proxy): scope the error-frame failure log to streams closed at the frame
A stream that ran to its end at the provider's error frame kept logging
through the success route before this PR, and the attempt-scoped failure
log is now taken only for a stream the router closed at that frame, with
the copy made at the close so the retry's deployment never reaches it
* chore: drop duplicate litellm import and mark the stream_ended rebind
* fix(proxy): log only router-superseded attempts as error-frame failures
A client that hangs up at a forwarded error frame closed the stream the same way the router closes a superseded attempt, so a non-retried stream lost its spend. The router now closes a superseded attempt through aclose_superseded_attempt, and only that close routes the attempt to the failure log
* test: type the _settle helper's recorded events
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* feat(ui): add the 1 hour cache write cost to the add model and model edit forms
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(e2e): cover both cache write tiers in the add model and edit forms
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(ui): hoist the expected pricing payload out of the inline assertion
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* refactor(ui): share the per-token conversion between both cache write tiers
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* refactor(ui): derive the touched pricing field type from the pricing field list
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* refactor(ui): collapse the duplicated pricing source keys into one map
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(ui): convert add model pricing once for all selected models and describe the 1 hour fallback
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(e2e): type the cache write tier deployment read back and fill the 1 hour field
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* style(e2e): keep the cache write tier spec on the dashboard prettier config
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
A native Responses WebSocket turn is now logged as an async call, so a proxy with a sync success
callback such as the response cache no longer runs each CustomLogger's sync success hook on top of
the async one. That second run counted every turn's tokens twice toward deployment TPM.
A turn whose admission check raises an unexpected error now gets a 500 error frame, the socket
closes with 1011, and the error becomes the socket's failure, on both the native and the managed
relay, instead of the socket dropping with no frame and no failure row.
A refused turn's failure row now carries the request's tags from the router's metadata, the same
way it already carries the model group.
* fix(proxy): streaming responses hand the event loop back every 50 ms instead of holding it for the whole buffered stream
* fix(proxy): yield to the event loop on every streamed chunk, no hold budget or clock
* docs(proxy): say why the streaming loop yields on every chunk
* fix(proxy): streaming loop yields once it has held the event loop 50 ms, clock read inline
* docs(proxy): streaming loop comment names the 50 ms hold
* Revert "docs(proxy): streaming loop comment names the 50 ms hold"
This reverts commit cbf1371873.
* refactor(proxy): streaming loop hold budget is _STREAM_LOOP_HOLD_MS = 50
* Revert "refactor(proxy): streaming loop hold budget is _STREAM_LOOP_HOLD_MS = 50"
This reverts commit 356454def6.
* docs(proxy): streaming loop comment notes the yield is a floor, not a count of releases
---------
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(count_tokens): apply context_management edits in the Gemini /v1/messages/count_tokens count
/v1/messages runs clear_tool_uses context edits before it sends an Anthropic
body to a Gemini or Vertex AI deployment, but /v1/messages/count_tokens counted
the body unedited, so a request with older tool results cleared counted about
2.8x what inference bills.
The count now runs the same history and context-edit step inference runs,
honors additional_drop_params on the deployment, and skips compact edits so a
count never makes a paid summary call.
* fix(count_tokens): match the Gemini count to inference for compact edits and drop opt-outs
Run compact edits in the count without the summary call so trigger checks and
history slicing match /v1/messages, honor a request-level additional_drop_params,
keep the full body when any deployment in the model group drops context_management,
and answer 400 naming a block Gemini cannot count when an edit trigger fails on it
* fix(count_tokens): count the unedited body when an edit fails and honor deployment ids in the drop check
* fix(count_tokens): log a failed edit in the count at debug, since the send already logs it
* fix(count_tokens): resolve a deployment id before wildcard routes in the drop check, as the router does
* fix(count_tokens): honor the router-wide additional_drop_params opt-out in the drop check
The router merges default_litellm_params between the deployment and the request, so a router-wide context_management opt-out makes /v1/messages send the full history. The count now reads it in the same order. Adds the audit's wire tests for the count's context edits.
* fix(count_tokens): read the opt-out of a deployment the router serves by its litellm model name
* fix(count_tokens): read the opt-out and drop_params from the deployments the router and completion_model send to
* fix(messages): drop an invalid context_management spec under drop_params so the send matches the count
* fix(count_tokens): give Claude Code and Codex counts the drop_params default /v1/messages gives them
The proxy turns on drop_params for Claude Code and Codex user agents on /v1/messages. The count now applies the same default, so an invalid context_management spec drops on both endpoints instead of the count answering 400 while the send answers 200. The wire test also covers default_fallbacks, model_group_alias, and completion_model routes.
* test(count_tokens): cover a coding agent that turns drop_params off and stop two count tests running as Claude Code
* fix(responses): log and bill every round of non-stream MCP auto-execute
Both model rounds run as internal sub-calls on the request's logging object, and the outer call logs the response the client received with usage, cost, and cost breakdown summed across the rounds. The internal-call gate now also covers the sync success callbacks, so an inner round no longer takes their first-wins slot
* fix(caching): let the parent call log and bill an internal sub-call's cache hit
A round of non-stream MCP auto-execute that the response cache served fired the
cache-hit callbacks on the shared logging object, so the spend log kept round 1's
function_call row and dropped the billed row for the answer. Internal sub-call cache
hits now fire no callbacks and carry a zero cost, and the summed cost breakdown leaves
out rounds served from the cache
* fix(logging): keep a parent row unique when its response came from the response cache
A RAG query or an MCP turn whose final round is served from the cache hands back the cached response's id, which already belongs to the row that cached it, so SpendLogs dropped the new row as a duplicate. Such a row now gets the cache-hit id suffix, and it is logged as a cache hit when it cost nothing.
* fix(logging): skip the zero-cost warning for a response served from the cache at no charge
A non-stream MCP auto-execute request whose rounds the response cache served is priced
at $0 before the parent marks it a cache hit, so it logged a false 'priced at $0'
warning and counted in litellm_zero_cost_requests_total. The diagnostic now treats a
response the cache served at no charge the same as a marked cache hit
* fix(responses): bill the paid first MCP round when a later step of the request fails
* fix(responses): log each MCP tool call under its own id so a failed follow-up keeps its spend row
The non-stream auto-execute path passed the request's call id to the tool call, so the tool's spend row and the request's failure row shared one request_id and the failure row was dropped on insert, leaving the paid first round unbilled.
* fix(responses): bill round 1 on a client disconnect, keep MCP tool rows in the request's session, and charge a fixed margin once per request
* fix(responses): keep bridged MCP auto-execute in one session and trace and continue its conversation
The bridged follow-up no longer chains on round 1, so a thinking model keeps its reasoning and the request stays in one session. The tool call log continues the request trace without renaming it, the tools are fetched once, and a later turn finds the MCP turn under the id the client got, skipping the tool rows.
* fix(responses): log the non-stream MCP tool listing in the request's session and trace
* fix(responses): find a bridged MCP turn by the id the client holds and log its listing in the bridged trace
* fix(responses): derive the MCP tool row id from the request's call id
The tool call's spend row now uses <call id>_<tool call id>, so it stays
linked to the request without colliding with the request's failure row.
Also tighten the internal cache-hit test to a fixed clock and typed params.
* fix(responses): bill a retried MCP attempt for its own rounds
A router retry or fallback reuses the request's logging object, so a failed attempt's stashed round 1 usage made the next attempt's success row log stale tokens. Each non-stream MCP attempt now starts without that stash.
* fix(responses): bill every failed MCP attempt's paid first round on the request failure
A retry or fallback now sets a failed attempt's stashed round 1 aside instead of dropping it, and the proxy's failure row adds it to the last attempt's own, so a request whose later attempt fails on its first call still bills the round the provider charged. A cached final round also keeps the paid first round's cost breakdown.
* fix(responses): keep MCP tool catalog and results out of the bridged next-turn replay
* fix(responses): keep MCP tool spend, client body keys, and Langfuse traces working on auto-execute
Count the MCP tools an auto-executed Responses API call ran in the tool spend rollup, accept a client body key named previous_response_client_ids without a 500, and point MCP listing and tool spans at the trace Langfuse gives the parent request, including x-litellm-session-id requests
* fix(responses): join MCP spans to the Langfuse trace when the request sends body metadata
* test(integration): cover non-stream MCP auto-execute billing, failure rows, and bridged follow-ups
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* fix(proxy): delete callbacks configured under callbacks or failure_callback
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(e2e): assert on configured callback rows and use a callback absent from CI configs
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(e2e): follow logging suite conventions for callback delete tests
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): flatten remaining callbacks with a helper generator
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): type callback delete helpers to satisfy basedpyright budget
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): save callback config without mutating the loaded dict
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): delete callbacks from every litellm_settings list GET shows
POST /config/callback/delete only searched success_callback while
GET /get/config/callbacks also lists failure_callback and callbacks, so
the Admin UI Delete returned 404 for those rows
Remove the name from every list that holds it, matching entries
case-insensitively, and skip lists the config file owns when the
database owns another copy so the save is not refused
Replace the e2e test with an integration test: the shared e2e stack
config owns litellm_settings.callbacks, so it cannot exercise that list
* test(proxy): cover config-owned success_callback on callback delete and wait on runtime state
Add the mirror mixed-ownership case where the config file sets
success_callback and the database owns failure_callback, which main
refused with a 400, plus a string-valued list case
Poll GET listings and /active/callbacks instead of reading them once, so
a request served by a worker that has not synced yet cannot fail the
test, and reuse the existing callback list key map in the route
* test(proxy): annotate callback delete test helper locals with Final
* fix(proxy): save callback delete through the config it read so concurrent settings edits are kept
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: yucheng-berri <293926326+yucheng-berri@users.noreply.github.com>
* fix(router): stop queueing budget increments for Redis when no Redis is configured
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(proxy): assert Redis-less model budget limiter tracks spend locally without queueing
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(responses): scope session history lookup to the referenced key
* Implement SQLite-backed DB for session tests
Added a SQLite-backed database class for testing and a new test to ensure session isolation between different API keys.
* fix(proxy): honor caller litellm_session_id in spend log session_id
Updated docstring to clarify session ID resolution logic for spend log rows.
* test(proxy): caller litellm_session_id wins over the trace id
* style(proxy): fix blank lines around new tests
* fix(proxy): read the caller session from the logging payload for spend logs
* fix(responses): match user and team too when replaying session history
* fix(proxy): keep bridged Responses chains whole and scope session history by owner
* fix(proxy): keep a metadata session next to a trace header and let session headers win
* fix(proxy): send the trace-header session only to spend logs and let session headers win beside a trace header
* fix(proxy): drop a client-sent caller session and keep it out of requester metadata
* fix(proxy): keep the trace id as the spend log session when the caller session contains a NUL byte
* fix(responses): keep a continued bridged chain's session under request_correlation_in_logs
With the flag on, the caller session outranked the chain session the bridge sets for a continued Responses chain, so a chain whose session changed or first appeared on a later turn split. The continuation now logs under the chain's session and drops the turn's caller session key, as on main. Adds the /audit integration cells for the spend log caller session.
* fix(responses): log a continued bridged chain under its session through the spend-log key only
The previous commit set the chain session as the turn's litellm_session_id, which also moved the OTel conversation id of a continued turn away from its trace header. The bridge now writes the chain session into the internal spend-log session key of the metadata the spend log reads, so only the spend log session changes, as on main. Adds an OTel cell for a continued chain.
* test(spend): cover bridged chains that share one caller session
* test(spend): cover the caller session on configured pass-through routes
* test(spend): type the SQLite fake and split stacked comprehensions in the caller session tests
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* fix(interactions): keep background interaction bills through crashes, late restarts, and deployment keys
A settlement claim is now a lease, so a row whose settler died between claim and
outcome is claimable again after the lease and boot resumes it. A resumed poll
gets the full poll window and fetches with the serving deployment's credentials
from the live router. The stored create context keeps the request tags and the
request metadata that the creating worker's own spend row used
* fix(interactions): bill a lapsed claim's takeover only when no spend row exists
A settler that billed and then died, or failed to record the outcome, left the row unsettled, so the replica that took over its lapsed claim billed the interaction a second time. A takeover now looks up the spend row keyed by the interaction id and records the row as billed when it finds one.
* fix(interactions): settle Vertex AI and edited deployments, never double bill with spend logs off
A takeover with spend logs off records the interaction unsettled instead of
billing it again, and a release records its outcome before freeing the budget
reservation. A resumed or live fetch carries the deployment's Vertex AI project,
location, and credentials, and a stale deployment id falls back to the model
group only when one same-provider deployment serves it. A poll stops at the
first 400, 401, 403, or 404, string request tags no longer fail the create,
and settled_at gets an index built concurrently.
* fix(interactions): leave a live replica's interaction to it when a resumed poll cannot read it
A replica that boots with credentials unable to read an interaction got a 404 on its resumed poll and settled the row as unsettled at once, so the replica that created it, still polling, found it settled and never billed it. A resumed poll that gets a 400, 401, 403, or 404 now stops fetching and keeps checking the row until its poll window ends, and settles it as unsettled only when no other replica did.
* fix(interactions): keep the creator's metadata on a resumed bill and retry outcome writes inside the claim lease
A bill another replica writes now carries the metadata the creating worker's own
spend row is built from, so a client's metadata.model_group no longer relabels it.
A failed outcome write is retried for half the claim lease before the budget
reservation is left to a takeover, and the creating replica retries every fetch
error for its whole window again, as main does. The resume integration case now
checks that a resumed fetch uses the deployment's own key.
* fix(interactions): resume background interaction polls only after database models load
Boot resume ran before the proxy loaded the models stored in the database,
so a resumed poll for a model added through /model/new or the Admin UI
fetched with env credentials. A 4xx there stops a resumed poll, so the bill
was lost. The settlement now installs after the database models load.
The integration test resumes a database model with an eager first fetch
and fails with the old startup order.
* fix(interactions): renew a settling claim so a slow settle is never taken over and billed twice
A settle runs every success callback in turn, and one slow callback could
hold it past the 5 minute claim lease. A booting replica or a DELETE then
took the live claim over, found no spend row yet, and billed the
interaction a second time. The settler now renews its claim every third
of the lease until it records the outcome, so only a settler that stopped
renewing is taken over.
A takeover also read the spend log after writing its claim, so a failed
read left the row under the taker's own fresh claim, which its poll then
took for another settler's and left unbilled until a reboot. The spend
row is now read before the claim is written, so a failed read takes
nothing and the claim is retried on the next poll.
* test(spend): cover lapsed claim takeovers, rotated deployment keys, late restarts, and resumed bill tags
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
* fix(proxy): count prompt-only token_counter requests with the provider when call_endpoint is set
* fix(proxy): keep the local count for a prompt-only token_counter request when the provider count fails
* fix(proxy): log a prompt-only token count fallback as a one-line warning
* fix(proxy): send a prompt-only token_counter request to the provider only on Gemini and Vertex AI
* fix(proxy): count a token_counter prompt sent next to an empty messages or contents list
* fix(proxy): count a prompt next to messages on the local token count instead of answering 500
* fix(proxy): answer 400 for a token count prompt next to a conversation unless drop_params is set
With drop_params the conversation is counted without the prompt. On the local count, a prompt sent with system or tools is counted as a user message with them, like the messages form.
* test(proxy): count tools with a prompt in the token counter fallback test
* test(proxy): cover prompt counting on /utils/token_counter in the Gemini count wire tests
* test(proxy): name the token counter fallback test by what it checks
* fix(proxy): track and drain management audit log writes on shutdown
* fix(proxy): track hook spawns and defer registration in audit log drain
* fix: satisfy LIT006 on audit log task registry by documenting the required cast
* fix(proxy): satisfy LIT006 and LIT010 in audit task registry
* test(proxy): drive the audit drain tests with events instead of wall-clock waits
* fix(proxy): drain audit writes beside the shutdown flushes and track only premium writes
Run the audit drain concurrently with the gateway and request-error flushes so
the two waits overlap inside the teardown budget, register the running audit
task only after the premium check, drop the four duplicate task-failure
loggers, and cover the shutdown order and a real hook spawn with tests
* fix(proxy): start hook audit writes before their email and secret manager calls
Key and user hooks now spawn their tracked audit write first, so a slow
invite email or secret manager call no longer spends the drain budget
before the write starts. The endpoints stop tracking the hook tasks, so
shutdown never waits on hook I/O when audit logging is off.
Adds regression tests for the hook order, the failed block or unblock
update, and the shutdown drain without a database.
* fix(proxy): track only the audit insert, not the code that awaits it
The shutdown drain now waits on each audit row insert itself, so it never waits on a caller's later work and covers every caller without per-endpoint wrappers. The user/new hook still tracks its own task because it reads the user row before the insert
* test(proxy): drain the audit registry in the config audit tests
The audit insert now runs in its own tracked task, one hop after the caller's task, so a single yield no longer reaches it
* fix(proxy): name dropped audit rows in the drain timeout warning
Each tracked audit task is named after its action, table, object id and destination, so a shutdown that drops a row says which one. Adds tests that the drain waits for slow audit callbacks and that the warning names the dropped row
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
