Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.orinadus.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Synopsis

urchin collect <COLLECTOR> [OPTIONS]

​
Subcommands

​
shell

urchin collect shell
Tails ~/.bash_history from the last recorded byte offset. New lines are ingested as command events with source: "shell". A checkpoint file at ~/.local/share/urchin/shell.checkpoint tracks the offset so reruns are incremental. No-op if ~/.bash_history does not exist.

​
git

urchin collect git --repo <PATH> [--repo <PATH> ...]
​
--repo
string
required
Absolute path to a git repository root. Repeatable. Can also be supplied via the URCHIN_REPO_ROOTS environment variable (colon-separated paths).
Reads new commits since the last recorded SHA for each repo. Ingests each commit as a commit event with the commit message as content, commit SHA in title, and source: "git". Per-repo checkpoints are stored in ~/.local/share/urchin/git-<hash>.checkpoint. First run per repo is silent — it records the current HEAD but does not ingest historical commits. 
urchin collect git --repo /home/user/dev/myproject

# Or via env var:
export URCHIN_REPO_ROOTS=/home/user/dev/myproject:/home/user/dev/other
urchin collect git

​
claude

urchin collect claude
Reads ~/.claude/projects/ JSONL transcript files. Ingests new conversation turns as conversation events with source: "claude". Uses file + line-offset checkpoints.

​
copilot

urchin collect copilot
Reads ~/.copilot/command-history-state.json. Uses a content-addressed hash checkpoint — reruns only ingest entries that weren’t present in the previous run.

​
gemini

urchin collect gemini
Reads ~/.gemini/tmp/*/chats/*.jsonl. Partial-offset checkpoint per file.

​
codex

urchin collect codex
Reads ~/.codex/state_5.sqlite, threads table. Captures first_user_message as intent. Row ID watermark checkpoint.

​
opencode

urchin collect opencode
Reads ~/.local/share/opencode/opencode.db. JOINs messages with sessions, filters to user-role messages. Row ID watermark checkpoint.

​
local-model

urchin collect local-model
Reads ~/.local/share/urchin/local-model.jsonl. Any local inference harness (Ollama, LM Studio, llama.cpp) can push events by appending newline-delimited JSON to this file: 
{"prompt":"fix the memory leak","model":"ollama:mistral","ts":"2026-05-01T10:00:00Z","workspace":"/opt/project"}
Fields: prompt (required), model (optional), ts (RFC3339, optional), workspace (optional). Urchin reads from this file and never writes to it. No-op when the file does not exist.

​
all

urchin collect all
Runs every collector that has a discoverable default path: shell, git (via URCHIN_REPO_ROOTS), claude, copilot, gemini, codex, opencode, local-model. Collectors whose source path does not exist are silently skipped.

​
Notes

  • All collectors are read-only. They never write back to source tools.
  • Each collector is idempotent — running it twice produces no duplicate events.
  • Collector errors are printed to stderr but do not affect other collectors when running all.