Introduction

reza is a local-first Universal LLM Context Database for coding agents.

The Problem

Every time you start a new AI session, you waste 5–15 minutes re-explaining your project:

  • What the stack is
  • Where the key files are
  • What was already tried
  • Why certain decisions were made

Switch from Claude to Cursor mid-task? Start over. Hit a context limit? Your architectural decisions vanish. Use two AI tools at once? They have no idea what each other did.

This is a solved problem. reza solves it.


The Solution

reza creates a local SQLite database (.reza/context.db) in your project that stores:

  • Every file path, type, line count, and purpose (extracted from docstrings and comments)
  • All active LLM sessions and their progress
  • Compact session summaries written by the active tool on save/end
  • Full raw conversation turns, searchable across the session with SQLite FTS5
  • A real-time change log synced via file watcher and git hooks
  • Transcript drops in .reza/handoffs/ for tools that export chats instead of writing turns directly

Any AI tool can query this database instead of scanning your files.

Current memory status

AreaStatus
Project memory.reza/context.db stores files, sessions, turns, FTS search, sources, checkpoints, threads, and privacy rules
Direct transcript syncCursor, Codex, and Aider can sync local transcript/history files through reza sync-all
Watch modereza watch updates project files, ingests .reza/handoffs/, and periodically runs configured conversation adapters
Thread continuityreza thread ..., reza session handoff --thread ..., and reza session search --thread ... are available
PC-wide registry~/.reza/registry.db tracks registered projects for reza global ... commands
Fallback ingestionGUI or unsupported tools can drop .md or .json exports into .reza/handoffs/
PrivacyBuilt-in redaction covers common assignment-style secrets and token formats before storage

Local recording verification

Use these commands on a project after installing Reza to prove that local agent history is being captured, indexed, and available to other tools:

reza install-hooks
reza sync-all
reza session list --json
reza session search "phrase from this chat" --json
reza context current --budget 8000
reza global status --json

Expected result:

  • .reza/adapters.json lists detected local tools and transcript sources.
  • .reza/context.db contains imported sessions and conversation_turns rows.
  • reza session search ... returns matching prior user/assistant turns with source_tool, source_path, and thread_id.
  • reza context current ... returns a compact thread-aware context packet that any coding agent can paste or fetch.
  • ~/.reza/registry.db contains this project so reza global search ... can route across the PC.

For tools without a stable local transcript path, export the chat into .reza/handoffs/ and keep reza watch running.