Claude Code
Install reza as a native Claude Code skill. Auto-triggers on session start and auto-syncs every response via Stop hook.
reza — Universal LLM Context Database
reza gives you instant project awareness without scanning files. Always query before touching code.
Step 1 — Check if reza is initialized
reza statusIf it fails: run reza init first.
Step 2 — Get project overview
reza queryThis returns: project name, language, framework, active sessions, and a breakdown of all indexed files with their purposes.
Step 3 — Find specific files
reza query --find "authentication"
reza query --find "api"
reza query --find "database"Use this instead of glob/grep when looking for files by purpose.
Step 4 — Check for interrupted sessions (handoff)
reza session handoffIf any sessions appear, use the summary and recent turns first. If you need older, specific context, search the raw transcript:
reza session search "keyword from the earlier discussion"Step 5 — Start your own session
reza session start --llm claude --task "describe what you are doing"Copy the session ID printed. Use it to save progress:
reza session save --id claude-XXXXXXXX \
--summary "what was accomplished" \
--context "key decisions, what to do next, what failed" \
--files "src/auth.py, src/models.py"Step 5b — Install the Stop hook (one-time, recommended)
reza install-claude-hookThis writes a Stop hook to ~/.claude/settings.json. After every response, Claude Code
reads its .jsonl conversation file and syncs new turns to reza automatically.
Zero tokens consumed. Works even when Claude hits its context limit.
You never need to call reza session turns add manually again.
Step 6 — End your session
reza session end --id claude-XXXXXXXX --summary "final summary"Query reference
reza query # full project overview
reza query --find TEXT # search by path or purpose
reza query --recent # last 30 file changes
reza query --sessions # active / interrupted sessions
reza query --file src/foo.py # info about one file
reza query --json # machine-readable outputExport (for non-SQL tools)
reza export # .reza/CONTEXT.md (markdown)
reza export --format json # .reza/context.json
reza export --format context # compact LLM prompt formatWorkflow rules
- Query first, code second. Never start reading files before running
reza query. - Use
--findinstead of glob.reza query --find authis faster and more accurate than globbing. - Install the Stop hook once.
reza install-claude-hook— turns are then saved automatically after every response, including at context limit. - Check handoff at session start. Always run
reza session handoff— if a previous session exists, its full turn history is ready to paste. - Search old chat when needed. Use
reza session searchinstead of relying only on recency. - Save a summary when switching tools.
reza session save --summary "..."gives the next tool human-readable context in addition to the raw turns.
File info
reza query --file src/api.py # shows purpose, line count, recent changesToken savings
reza querycosts ~200 tokens vs ~18,000 tokens for manual file scanningreza session handoffcosts ~300 tokens vs 10,000+ tokens to re-explain context- Use reza to stay within context limits on large projects