Give Claude Code, Codex, and other agents the codebase knowledge your team spent years building.
- •Analyzes your commit history
- •Tracks every function call
- •Indexes all your tests
Codeset generates the repository context your coding agents need to understand your codebase.
$5 per repo, one-time. No subscription. Ready in ~30 minutes.
With the information provided by Codeset, Haiku 4.5 delivers better performance than baseline Sonnet 4.5 and Opus 4.5, significantly reducing costs.
Read the full evaluation →
#847 · your-org/analytics
OpenFix: dashboard charts show flat lines for all metrics since last Tuesday.
Coding Agent· Retrieves Codeset-generated context
get_context.py src/dashboard.pysrc/dashboard.pysrc/data_pipeline.py [co-change]Coding Agent· Reasoning
Related file note shows migration #412 deprecated event_count — column still exists but is no longer written to. get_chart_data() still queries the old column, returning zeroes since the migration ran. Switching to events_total.
Coding Agent· Fix Applied
def get_chart_data(metric, date_range): return db.query(- "SELECT date, event_count "+ "SELECT date, events_total " "FROM daily_stats " "WHERE date BETWEEN %s AND %s", date_range )
The agent queries our knowledge base, reasons with it, and finds the correct fix.
Conventions and architecture are easy.
The harder knowledge is implicit.
Which files tend to break together. Which functions have been misused before. Which edge cases burned your team three months ago and are about to happen again.
That knowledge lives in your git history, your test suite, and your AST — not in any README. We extract it automatically and surface it to your agent at the moment it matters.
What we extract
- →Commit history insights. Past bugs, root causes, failed attempts, and lessons learned — structured and linked to the files they affected.
- →Per-file pitfalls. Specific relevant mistakes each file has caused, the consequence, and the prevention. Extracted from your history, not invented.
- →Test coverage map. Which test files and functions exercise each source file. Your agent knows exactly where to look after making a change.
- →Co-change patterns. Files historically modified together, exposing hidden coupling that imports alone don't reveal.
- →AST caller graph. For every function, we know every caller: file, line number, and call context. Your agent understands impact before touching a line.
What you get — for $5
A per-file knowledge base your agent can query as it works, and improved CLAUDE.md/AGENTS.md files.
AGENTS.md / CLAUDE.md
Improved AGENTS.md/CLAUDE.md providing a high-level overview of your codebase.
# CLAUDE.md — acme/webapp ## Architecture src/api/ # React Query hooks + axios clients src/stores/ # Zustand — one store per domain src/pages/ # Route-level components (flat) src/components/ # Atomic design UI library ## Key Commands pnpm test:unit # Vitest + Testing Library pnpm test:e2e # Playwright pnpm lint:fix # ESLint + Prettier ## Conventions - Functional components only - Named exports preferred - Mock src/api/* in every unit test ## Gotchas Zustand uses the slice pattern. See src/stores/README first. Never call store actions during SSR.
Per-file knowledge base
For every file in your repo: past bugs with root causes, specific pitfalls with consequences, every function caller with line numbers, and which tests exercise it. Queryable by your agent on demand.
$ python get_context.py src/payments.ts
── src/payments.ts ──
History (4 insights):
[Bug Fix] Double-charge on retry
Root cause: idempotency key not persisted
Fix: store key in DB before Stripe call
[Breaking Change] Webhook signature v2
Migration: update STRIPE_WEBHOOK_SECRET
Pitfalls:
✗ Don't call charge() inside a DB transaction
→ Stripe call may succeed but rollback fires
✗ Never log the full PaymentIntent object
→ Contains raw card data (PCI violation)
Callers (6 files):
api/checkout.ts:88 handleCheckout()
api/subscriptions.ts:41 renewSubscription()
workers/retry.ts:15 retryFailedCharges()
... 3 more
Tests → tests/payments.test.ts:
test_successful_charge, test_idempotent_retry,
test_webhook_verification, test_refund_flow
Co-changes: src/webhooks.ts, src/subscriptions.tsWhat your agent sees when it reads src/payments.ts
- ›The double-charge bug that hit production — and the fix.
- ›Two pitfalls with consequences: don't call inside a transaction, never log the full object.
- ›Six callers across the codebase, with file and line.
- ›Four tests, ready to run after any change.
- ›Two files that always move with this one.
Generated automatically from your git history and AST. No manual annotation. No maintenance.
How it works
Connect your repo
Paste your GitHub URL and sign in. We support private and public repositories.
We analyze your codebase
Our pipeline mines your commit history, traces every function caller, extracts pitfalls, and maps test coverage. Under 30 minutes.
Download and commit
Get your AGENTS.md and per-file knowledge base. Commit them — your agent uses them from now on.
Common questions
Anything not covered here — reach out via the contact page.
Start with your most important repo.
$5 per repo. One-time — no subscription, no seat licenses. Your agent gets wired into your codebase's history in under an hour.
What you get
- ✓CLAUDE.md / AGENTS.md
- ✓Per-file knowledge base
- ✓Commit history analysis
- ✓Full AST caller graph
- ✓Test-to-file mapping