Skip to main content

Reporecall

Reporecall is the local-first, auto-injecting, self-aware-about-staleness context + memory layer for coding agents.

It automatically injects high-quality, compressed evidence (code + wiki + graph + business + memory) into every Claude Code prompt via hooks, while being brutally honest about freshness.

It is a context layer — not an AI editor or hosted model. Its job is to make the context your agent receives dramatically better, especially on large, high-churn, or unfamiliar codebases.

 ____                                    _ _
|  _ \ ___ _ __   ___  _ __ ___  ___ __ _| | |
| |_) / _ \ '_ \ / _ \| '__/ _ \/ __/ _` | | |
|  _ <  __/ |_) | (_) | | |  __/ (_| (_| | | |
|_| \_\___| .__/ \___/|_|  \___|\___\__,_|_|_|
          |_|

The problem it solves​

Coding agents work best when the right files are already in context. Left on their own they burn tokens grepping, re-reading files, and guessing at architecture — and they still miss cross-module relationships that only a call/import graph can reveal.

Reporecall front-loads that work. It maintains a fresh index of your code, understands what a prompt is really asking (an exact lookup vs. a trace vs. an architecture inventory), and assembles a token-budgeted bundle of the files and symbols that actually matter — including graph neighbors the agent would not have found by keyword search alone.

Key features​

  • Auto-injecting hooks (the killer UX). Relevant context is pushed into every prompt via Claude Code hooks. The agent doesn't have to decide to call a tool.
  • Explicit Trust Contract. Every response carries staleness banners, indexedCommit, dirty file count, and repair guidance. Never silently stale.
  • Intent-routed hybrid retrieval + compression. Routes lookup/trace/bug/architecture/change/skip, then compresses with expand-on-demand (search_code action=read_chunk).
  • Full local bundle. Code + call graph + deterministic wiki + persistent memory + business areas + Lens — zero external services required by default.
  • Six-tool MCP surface. Deliberately compact and safe (search_context, search_code, explain_flow, memory, refresh_context, get_stats).
  • Architecture Lens + exports. Portable HTML dashboard + rich JSON for other tools.

How it works (high level)​

flowchart TB
  Q["User or agent question"]
  Entry["Hook, CLI, MCP, or JSON command"]
  Intent["Intent classifier"]
  Search["Code retrieval (keyword + vector, RRF)"]
  Graph["Graph expansion (calls / imports)"]
  Wiki["Wiki + product-area evidence"]
  Memory["Project memory"]
  Selected["Assembled, token-budgeted context"]
  Agent["Agent reads selected files first"]

  Q --> Entry --> Intent
  Intent --> Search --> Graph --> Selected
  Intent --> Wiki --> Selected
  Intent --> Memory --> Selected
  Selected --> Agent
  1. Index. Tree-sitter parses source into chunks; imports and call edges are recorded; metadata goes to SQLite, vectors to LanceDB.
  2. Classify. The prompt is sanitized and routed to an intent mode.
  3. Retrieve. Hybrid search finds candidate chunks; the graph adds callers, callees, and import neighbors.
  4. Assemble. Context is packed under a token budget, with secondary evidence compressed (and reversible via search_code action=read_chunk).
  5. Deliver. The bundle is returned through Claude Code hooks, MCP tools, or CLI explain/search.

See Architecture for the full pipeline.

Why local-first​

Reporecall runs entirely on your machine. The default retrieval backend uses local SQLite/FTS indexes and local vector embeddings (Xenova/all-MiniLM-L6-v2); no cloud embeddings API is required. That means:

  • Privacy. Your source never leaves the machine unless you explicitly choose the openai embedding provider.
  • No network dependency. Indexing and retrieval work offline.
  • Deterministic outputs. Wiki pages and the lens JSON are generated deterministically from the index, so external tools can rely on them.

Optional semantic backends (ollama, openai) can be configured separately if you want them — but the out-of-the-box experience needs no external services.

Where to go next​