Skip to content

feat(review): raise the RAG per-repo chunk cap from 1500 to 4000 #4398

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
JSONbored merged 1 commit into from
Jul 9, 2026
Merged

feat(review): raise the RAG per-repo chunk cap from 1500 to 4000 #4398

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 10 additions & 2 deletions src/review/rag.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,8 +100,16 @@ export const RAG_DIMENSIONS = 1024;

const CHUNK_CHARS = 16000; // per-file chunk budget; only files larger than this are split
const CHUNK_OVERLAP = 1500;
/** Hard per-repo stored-vector cap — the free-tier guard. Source is prioritized so it survives the cap. */
export const MAX_CHUNKS_PER_REPO = 1500;
/** Hard per-repo stored-vector cap — bounds a repo-controlled, unboundedly-growable store. Source is
* prioritized so it survives the cap. Raised from 1500 (2026-07-09): all 3 currently-gated repos were
* sitting AT the old cap (confirmed live), and gittensory's own indexable tree alone is within range of
* it before accounting for large files splitting into multiple chunks -- meaning real code was silently
* never indexed. Storage/search cost at this scale is trivial for Qdrant regardless of cap size (the
* bound exists to protect against a pathological future repo, not because 4000 vectors is expensive);
* the real cost of a higher cap is embedding COMPUTE, which is now a one-time cost per file instead of a
* recurring one per cron cycle (#4365's blob-SHA skip-cache). Self-host only: this is not a Cloudflare
* free-tier constraint on this deployment, but the name/comment history predates self-host. */
export const MAX_CHUNKS_PER_REPO = 4000;
const EMBED_BATCH = 96; // Workers AI caps embedding input at 100 items/call; kept as a conservative general
// bound — other embed providers (Ollama/vLLM/etc via the self-host adapter) may not share this exact cap.
const MAX_CONTEXT_CHARS = 14000; // bound the injected block (mirrors diff/knowledge budgets)
Expand Down