GitNexus/docs/wiki-generation.md
enih 1e5a1ad01e
fix(wiki): configure limits and retry truncated pages (#3775)
* feat(wiki): let callers set the completion token limit

Reasoning models can spend the hardcoded 16384 cap before a page is finished. Keep that default, and honor a flag, GITNEXUS_MAX_COMPLETION_TOKENS, or maxCompletionTokens in the CLI config.

Fixes #3774

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(wiki): reject LLM responses cut off at the completion token limit

callLLM accepted a response whose finish_reason was "length" as a
finished one, streamed or not, so a page cut off mid-sentence or
mid-table was written and the run reported success. With a reasoning
model the limit can be spent before the answer starts.

Both paths now throw when the provider reports finish_reason "length",
naming the limit. The generator already treats a thrown error as a
failed module: the page is not written, the module is listed under
"Failed modules", and the next run regenerates it.

---------

Co-authored-by: nullptr <null@null.ptr>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Samuel Mbabhazi <samuelmbabhazi@gmail.com>
Co-authored-by: Gergo Magyar <gergomagyar0@gmail.com>
Co-authored-by: Gergő Magyar <gergomagyar@icloud.com>
2026-10-31 13:24:37 +00:00

2.7 KiB

Wiki generation

Back to GitNexus

Generate LLM-powered documentation from your knowledge graph:

# Requires an LLM API key (OPENAI_API_KEY, etc.)
gitnexus wiki

# Use a custom model or provider (default model: minimax/minimax-m2.5)
gitnexus wiki --model gpt-4o
gitnexus wiki --base-url https://api.anthropic.com/v1
gitnexus wiki --provider grok   # local Grok Build CLI (uses `grok login`, no API key)

# Force full regeneration
gitnexus wiki --force

# Increase the timeout or retries for large codebases or slow LLM providers
gitnexus wiki --timeout <seconds>  # LLM request timeout in seconds (default: disabled)
gitnexus wiki --retries <n>        # Max LLM retry attempts per request (default: 3)

# Raise the completion-token limit (default: 16384)
gitnexus wiki --max-completion-tokens 131072
GITNEXUS_MAX_COMPLETION_TOKENS=131072 gitnexus wiki
# Or set "maxCompletionTokens": 131072 in ~/.gitnexus/config.json
# Precedence: flag, then the environment variable, then the config file.

# Allow a specific LAN/self-hosted HTTP LLM host (HTTPS is preferred for remote endpoints)
gitnexus wiki --base-url http://llama-box.local:8080/v1 --allow-insecure-connection llama-box.local
# Or set a comma-separated host allowlist:
GITNEXUS_ALLOW_INSECURE_CONNECTION=llama-box.local,192.168.1.23

# Change the output language
gitnexus wiki --lang <lang>  # e.g. english, chinese, spanish, japanese

The completion limit must be a positive integer supported by your model. It applies to OpenAI-compatible HTTP providers; local agent CLI providers manage their own limits. For reasoning models, the limit includes reasoning tokens as well as the visible answer. When a provider reports finish_reason: "length", GitNexus rejects the incomplete response instead of saving it as a finished page. Failed modules are listed, the command exits with status 1, and a normal rerun retries them even if the repository commit has not changed. Increase --max-completion-tokens within your model's supported limit before retrying. Truncated grouping or overview responses also fail the run; batched grouping retains its directory-based fallback. Incomplete runs are not published with --gist. Use --force to replace incomplete pages already saved by an older version.

For safety, http:// LLM base URLs are allowed by default only for loopback hosts (localhost, 127.0.0.1, ::1). --allow-insecure-connection and GITNEXUS_ALLOW_INSECURE_CONNECTION accept exact hostnames or IP addresses only; do not include schemes, ports, paths, credentials, or wildcards.

The wiki generator reads the indexed graph structure, groups files into modules via LLM, generates per-module documentation pages, and creates an overview page — all with cross-references to the knowledge graph.