Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/self-hosting/govern/plane-ai/custom-llm.md`:
- Line 95: The long documentation sentence "Plane AI activates inference profile
mode automatically when a profile ARN or ID is set and ambient AWS credentials
are present (`AWS_ROLE_ARN`, `AWS_WEB_IDENTITY_TOKEN_FILE`,
`AWS_CONTAINER_CREDENTIALS_FULL_URI`, or
`AWS_CONTAINER_AUTHORIZATION_TOKEN_FILE`)." exceeds 120 characters; break it
into shorter lines (e.g., two sentences or a short paragraph) or convert the
credential list into a bulleted list so no line exceeds 120 characters, update
the surrounding text to keep meaning unchanged, and run Prettier to ensure the
formatting check passes.
- Line 17: The sentence "Use a model with at least 1 trillion parameters" is
inconsistent with the examples (llama-3.3-70b-versatile and llama3); update that
requirement to the correct minimum (e.g., "70+ billion parameters" or "1
billion+" as appropriate for Plane AI features) so the guidance matches the
examples, and edit the sentence in custom-llm.md accordingly (referencing the
exact string "Use a model with at least 1 trillion parameters" and the example
model names "llama-3.3-70b-versatile" and "llama3" to locate and verify
consistency).
- Line 13: The table row under the "OpenAI-compatible" cell is over the 120-char
Prettier limit; shorten or reformat that cell to pass CI by either truncating
the provider list (e.g., "OpenAI-compatible (Ollama, Groq, vLLM, LiteLLM, ...)")
or breaking the content into multiple table rows or a wrapped multiline cell so
the line length is ≤120; update the cell text that currently reads "Your model
is served via an OpenAI Chat Completions API (Ollama, Groq, vLLM, LiteLLM,
Cerebras, etc.)" accordingly.
- Line 88: Add a brief clarifying comment next to the CUSTOM_LLM_MODEL_KEY entry
to explain that this value is the simplified inference profile ID (used for
cross-Region system-defined Bedrock inference profiles) and intentionally
differs from the full foundation model ID used with standard credentials;
reference the symbol CUSTOM_LLM_MODEL_KEY and include an example of the full
format (e.g., anthropic.claude-3-5-sonnet-20241022-v2:0) vs the simplified ID
(claude-sonnet-4-6) so readers understand the distinction.

In `@docs/self-hosting/govern/plane-ai/getting-started.md`:
- Line 11: Remove the duplicate sentence "For an overview of what Plane AI can
do, see [Plane AI](https://docs.plane.so/ai/pi-chat)." so it appears only once
in the document; locate both occurrences of that exact string and delete one of
them (either the instance at the earlier position or the later one) to eliminate
redundancy.
- Around line 26-27: Update the listed model versions in the getting-started
docs to remove non-existent entries and replace them with the currently
available recommended versions: remove "GPT-5.2" and any mention of "Claude
Sonnet 4.5", and add OpenAI models "GPT-5.5", "GPT-5.4", "GPT-5.4-mini",
"GPT-5.4-nano" and Anthropic models "Claude Opus 4.7", "Claude Sonnet 4.6",
"Claude Haiku 4.5" where "GPT-5.2" and the old Sonnet entry appear so
configuration examples and defaults reference valid model IDs.

In `@docs/self-hosting/govern/plane-ai/semantic-search.md`:
- Line 32: The doc currently says a mismatch for OPENSEARCH_EMBEDDING_DIMENSION
"breaks indexing" but later notes it "breaks indexing silently"; make these
consistent by updating the earlier sentence that mentions
OPENSEARCH_EMBEDDING_DIMENSION so it reads that a mismatch "breaks indexing
silently" (or equivalent phrasing) and verify the later sentence uses the same
wording to avoid ambiguity.

---

Nitpick comments:
In `@docs/self-hosting/govern/plane-ai/getting-started.md`:
- Line 50: The sentence describing FOLLOWER_POSTGRES_URI is incomplete; update
the line for **`FOLLOWER_POSTGRES_URI`** to finish the thought (e.g., "Can be a
read replica of the primary database.") so it reads as a full sentence and
clearly explains that this URI may point to a read replica used for read-only
access to workspace data.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/self-hosting/govern/environment-variables.md`:
- Line 213: The docs show two different defaults for CUSTOM_LLM_MAX_TOKENS (one
entry lists 64000, another lists 128000); pick the correct canonical default and
update both occurrences so they match—specifically change the standalone
environment variable row describing CUSTOM_LLM_MAX_TOKENS and the Plane
Intelligence table entry that references CUSTOM_LLM_MAX_TOKENS to the same value
and wording (e.g., “Defaults to `64000`” or “Defaults to `128000`”) to eliminate
the conflict.