Best integration

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 status

If it fails: run reza init first.

Step 2 — Get project overview

reza query

This 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 handoff

If 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"
reza install-claude-hook

This 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 output

Export (for non-SQL tools)

reza export                   # .reza/CONTEXT.md (markdown)
reza export --format json     # .reza/context.json
reza export --format context  # compact LLM prompt format

Workflow rules

  1. Query first, code second. Never start reading files before running reza query.
  2. Use --find instead of glob. reza query --find auth is faster and more accurate than globbing.
  3. Install the Stop hook once. reza install-claude-hook — turns are then saved automatically after every response, including at context limit.
  4. Check handoff at session start. Always run reza session handoff — if a previous session exists, its full turn history is ready to paste.
  5. Search old chat when needed. Use reza session search instead of relying only on recency.
  6. 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 changes

Token savings

  • reza query costs ~200 tokens vs ~18,000 tokens for manual file scanning
  • reza session handoff costs ~300 tokens vs 10,000+ tokens to re-explain context
  • Use reza to stay within context limits on large projects