open source
Deterministic log templating: the base layer that turns raw log lines into template groups before any model runs.
A Drain-based grouper with a codag adaptation for compact, punctuation-heavy logs. Bounded raw samples and slot summaries, no inference, fully reproducible. This is the open engine under codag wrap.
$ codag-drain < app.log 1,240,000 lines 14 templates [812,394] GET /v1/users/<*> 200 [404,001] cache <*> user:<*> [ 23,118] pool acquire <*>ms [ 412] ERROR psycopg2.OperationalError
See your AI workflows as live diagrams, right inside VS Code. Works with any language, model, or framework.
Edit code → graph refreshes live. Click a node → jump to source. Tree-sitter parsing under the hood, no config required.
The open-source command-line client and MCP server for compressing logs before they hit an AI agent.
Wrap provider CLIs, pipe stdin, or register Codag tools with Claude Code, Codex, and Cursor. OAuth login, local config, and opt-out telemetry are documented in the repo.
$ codag wrap -- kubectl logs api -n prod --tail=2000 reading 2,000 log lines... codag output service: api pattern: database pool exhaustion evidence: repeated connection timeout next: inspect pool size and idle timeout