Every repository handed its `.table` back untyped, so a dozen modules had
each grown a private `_PrismaTableActions` Protocol to paper over it. They
had drifted: some declared `update` as returning the row, others the row or
None, and none agreed on whether `find_many` was covariant
Replace all of them with a single `TableActions[RowT_co]` in
`litellm/repositories/prisma_protocols.py`, keyed to the prisma row each
repository is bound to. Query inputs stay `Mapping[str, object]` so callers
keep passing plain dicts, and `find_many` returns `Sequence` so the row type
stays covariant
Typing the nullable returns honestly surfaced paths that were already
crashing. A team admin could never edit or delete a memory entry owned by
their team: the write-auth check fed a raw prisma row to a helper that
expects the domain model, so `members_with_roles` arrived as plain dicts and
the request died as a 500 instead of applying the edit. Non-admin members hit
the same 500 in place of the 403 they were owed, so refusal and breakage were
indistinguishable. `/v2/model/info?user_models_only=true` dereferenced a
missing user row rather than returning the 400 the route already had, three
team routes dereferenced a team deleted between the read and the write, and
the agent registry dereferenced a missing agent instead of naming it
basedpyright drops 2,132 errors, 1,454 of them reportAny and 73
reportExplicitAny. The dashboard's generated types pick up `string[]` where
they had `unknown[]` for a team's members, admins and models
The LLM classifier capped every prior turn at 200 characters independently, so a
785 character turn was cut even when the whole block it belonged to was 353
characters. A character budget now bounds the block: turns are taken newest first
and quoted whole while they fit, older turns are dropped whole once it runs out,
and only the turn straddling the boundary is cut. The per-turn cap stays as an
optional clamp for operators who set it deliberately, defaulting to unset.
Adds a scheduled GitHub Actions lane on top of the merged record/replay
transport. A Saturday cron records the `replayable` e2e tests against the
real providers and publishes the fixture bundle as a private
`e2e-fixtures-bundle` artifact with a SHA-256 sidecar. Weekday crons pull
that artifact by its pinned digest, verify the checksum before extracting,
and replay it with provider credentials set to bogus values, so a run that
ever reached a real provider fails instead of passing.
An egress sentinel pins the provider hostnames to a local sink for the whole
replay job and counts every connection that reaches them; the job asserts
that count is zero, so hermeticity is proven by measurement. A red Saturday
publishes no bundle, so the next weekday finds nothing fresh and fails loudly
rather than replaying a week-old recording, and the transport's seven-day
freshness gate hard-fails any bundle that has drifted too far. The lane also
runs on demand from the Actions tab with a record/replay `mode` input.
Tests join the lane with `@pytest.mark.replayable`. The streaming Anthropic
test now counts to twenty so its recorded response banks several content
deltas, matching the assertion that the stream arrives incrementally.
A request carrying a session/trace header fans the header value into
litellm metadata as both trace_id and session_id. LangSmith then rejected
the whole ingest batch: a root run's trace_id must equal the run id
embedded in dotted_order (400), and a run-body session_id must reference
an existing tracer session (404/422). Override caller trace_id on runs
that post as roots and drop session_id only when it mirrors trace_id,
so deliberate child-run and valid tracer-session fields still pass through.
The error file now resolves through _provider_output_file_id like the output
file does. Sending the encoded id straight to the provider 404s, and the
swallowed fetch failure would silently report zero failures.
Staging split batch output-line costing into _safe_output_line_stats /
_compute_output_line_stats / _output_line_cost so one uncostable line can no
longer zero a whole batch, and added _provider_output_file_id so model-encoded
output file ids decode before the fetch. This branch's pass/fail counting was
written against the pre-split shape, where every None line meant a provider
failure.
Keep staging's structure and layer the counts on a three-way classification: a
provider-reported failure yields PROVIDER_FAILED, a provider-successful line
litellm cannot price yields UNCOSTABLE and stays in successful_requests billed
at $0. Without that split a litellm-side pricing gap would be reported to the
customer as a failed request and the counts would stop reconciling with the
provider's own request_counts.
Route the error-file fetch through _provider_output_file_id too, and carry the
new dataclass return through the callers staging added after this branch
forked.
A quoted routine call qualified by a schema and sitting inside a CREATE INDEX
expression, ON "Foo" (public."f"(col)), walked its qualifier read-through back
across the opening paren to the ON that introduces the indexed table, so the call
was misread as a relation and dropped from the call set, leaving a rewrite in that
routine unscanned. A word now only introduces the name when nothing but whitespace
and qualifier dots lies between them, so a paren in that gap keeps ON (and any
relation-introducing keyword) from reaching across it and the call stays a call.
The Responses-API to /chat/completions bridge yields ModelResponseStream
chunks that carry choices followed by a trailing event object that has no
choices key. stream_chunk_builder assumed every chunk was subscriptable at
"choices", so assembling those chunks raised KeyError('choices') and was
re-wrapped as a 500 APIError building the streaming usage.
Guard each choices access with .get("choices") so choices-less chunks are
skipped instead of crashing. Behavior is unchanged for chunks that do carry
choices, since .get("choices") is truthy only for a non-empty choices list.
Adds a regression test that assembles content across chunks followed by a
trailing chunk with no choices key.
Co-authored-by: mubashir1osmani <mubashir.osmani777@gmail.com>
/global/activity/cache_hits now returns an error_breakdown: failed spend
logs bucketed per call_type by error code and error class, read from
metadata->error_information. Clicking a red failed-requests segment on
the cache activity chart opens a per-code bar chart; hovering a bar
lists the error classes behind that code.
The video edit endpoint parsed the multipart body but dropped the uploaded
source video, only normalizing it to an id. When a raw file is uploaded it now
flows through videos.main -> the http handler -> the provider transform, which
emits multipart/form-data with the source video as a file part, matching the
official OpenAI SDK's videos.edit wire format. Edit-by-id still egresses JSON.
OpenAI videos never fetches input URLs or invents download size caps. Drop MAX_VIDEO_MEDIA_* and the fetch-and-inline path. Keep MAX_IMAGE_URL_DOWNLOAD_SIZE_MB for chat image handling. Validate Omni reference URLs, then send them as form fields the way OpenAI videos sends input_reference as a file
The LLM classifier's conversation context cut each prior turn head-only, so a turn
opening with an incident report and closing with the actual request reached the
classifier as the incident report alone. Keeping head and tail costs the same
budget and is what the truncation literature measures as best for classifying
long text.
Suppress the three reportUnknownArgumentType diagnostics with reasons at
the untyped provider-params boundary, collapse the early return, and
assign extra_body via a TypedDict-annotated literal so the file's
basedpyright profile matches the merge base exactly. The user-supplied
extra_body merge is covered end-to-end through get_optional_params.
On mapped pass-through routes, of which /vertex_ai is one,
user_api_key_auth accepts the caller key from a header literally named
litellm_user_api_key and applies it last, so it overrides every other source.
The credential-less filter neither dropped it nor resolved the caller key from
it, so a virtual key there reached Google past a real x-goog-api-key, and a
bring-your-own Authorization could be stripped when auth actually came from that
header. Drop it by name and resolve it at highest precedence.
The recursive_detector code-quality gate fails on litellm_internal_staging
because _flatten_form_field and _flatten_form_data_field in
llm_request_utils.py are recursive but absent from IGNORE_FUNCTIONS. Both are
bounded structural recursion over an already-parsed JSON-shaped request body
(a finite tree, no cycles possible), matching the existing ignored walkers, so
add them to the ignore list with a justification comment.
Follow-up to #38130. The function has no callers in the repo or the docs and is
not exported from `litellm/__init__.py`, and `token_counter` already does the same
job better, so keeping a second entry point only preserves a trap.
That trap is real: Greptile flagged on #38130 that `token_counter` picks the claude
tokenizer only for bare ids. `claude-sonnet-4-5` resolves to huggingface_tokenizer,
while `claude-3-opus-20240229` and `anthropic/claude-sonnet-4-5` fall back to the
OpenAI one, 24 tokens against 27 on the same string. Deleting the wrapper removes
the surface rather than papering over it; the selection gap in `token_counter`
itself is worth its own fix.
BREAKING CHANGE: `from litellm.utils import prompt_token_calculator` no longer
resolves. Use `litellm.token_counter(model=..., text=...)`.
Route hosted_vllm video generation through /v1/videos as multipart form data so Omni extra fields such as width and extra_params reach the server instead of a JSON body Omni rejects
A quoted routine call with a SQL comment between its name and parenthesis,
`"backfill" /* reason */ ()`, is a real call that rewrites rows at boot, but the
call-site check read the raw SQL and stopped at the comment, so the routine was
read as uncalled and its rewrite slipped through. Read the call test from the
masked text instead, where every comment is already blanked to spaces, so a
comment between the name and its parenthesis is skipped exactly as whitespace is,
line, block and nested block comments alike, while a like-named non-call
identifier still opens no call and stays masked
The resolver placed both operator-configured key headers at the top of its
precedence, but user_api_key_auth only overrides with litellm_key_header_name;
a pass_through_endpoints litellm_user_api_key is checked last. So a request that
authenticated via Authorization while also sending a pass-through header could
have the wrong value chosen, leaving the authenticated Authorization key
forwarded. Order the resolver exactly like get_api_key: override first, built-in
headers next, pass-through header last.
A migration that defines an uncalled row-rewriting routine and elsewhere
references a quoted column, table, index, or constraint sharing the routine's
name was wrongly flagged: the guard restored every double-quoted identifier
before the name search, so a like-named identifier read as a call. Restore only
quoted names that open a call, followed by "(", so a routine invoked through a
quoted identifier is still caught while a like-named non-call identifier stays
masked and no rewrite-free migration is rejected
user_api_key_auth also accepts the caller key from a pass_through_endpoints
entry's headers.litellm_user_api_key, not just litellm_key_header_name. Drop
every operator-configured caller-key header by name and treat them as
top-precedence caller-key sources, so a virtual key sent through one is never
forwarded to Google.
