Show HN: Ambits – agentic grep/rg tool will history tracking

Ambits is a developer tool that gives AI coding agents symbol-aware grep/rg functionality plus a persistent journal of what the agent has already read. It exposes symbol definitions as structured JSON addressed by symbol id or content hash and records every symbol access with a read depth (overview vs full body) so a compaction-aware agent can restore prior context instead of re-parsing or re-reading files. Outputs are plain text or JSON, a live TUI is available, and Claude Code ingestion has been exercised. The show command returns exact source spans (or metadata-only), supports content-hash selectors, and reports ambiguity rather than attempting to resolve duplicate identifiers.

Search and caller-finding are integrated with tree-sitter: every grep hit is attributed to the innermost containing symbol and annotated with whether that symbol has been seen this session. ambits provides two frontends (rg and grep) to honor ripgrep and grep flag semantics, uses ripgrep’s regex engine and many of its flags, and deliberately deviates where agent contexts demand determinism and brevity (sorted output, column on by default, 200-match head-limit, 300-column clip). The pipeline scans raw bytes first and parses only matching files for speed. Callers are discovered via grammar tags (name-matching only, not type resolution), and references are computed on demand to minimize cost while giving agents accurate, compact code slices and a durable memory of what was examined.