REI Code

Claude Code–style terminal coding agent powered by REI Labs Units/Reigents.

What it does

• Interactive and one-shot CLI for repo-aware coding assistance
• Tooling: read/write/apply patch, search, git status/diff/log, run commands (deny-by-default)
• Session persistence and resumable conversations (repo-local state)
• REI Labs provider integration (Units/Reigents) with debug logging and retry/backoff (no streaming yet)
• REPL IO purity: all REPL/slash output flows through injectable IO channels (write/info/warn/error/debug/log)

Install (npm)

Global install (adds rei to your PATH):

npm install -g rei-code
rei --help

Run without global install:

npx rei-code --help

Prerequisites

• Node.js 18+ (required for CLI runtime)
• REI Labs API key (export REI_API_KEY or pass --api-key)

For contributors/developers in this repo:

• Bun 1.0+ (used for dev workflow and tests) — install from https://bun.sh

Dev install & build

bun install
bun run build

Run

Interactive REPL:

bun run start
# or
node dist/index.js

One-shot prompt:

node dist/index.js "Add tests for foo"

Print/pipe mode (Claude parity flag):

echo "Summarize src/cli.ts" | node dist/index.js -p

Common flags (parity-inspired):

• -p, --print – print response and exit
• -c, --continue – continue most recent session in this repo
• -r, --resume <id> – resume by session id
• --unit / --reigent <name> – pick model/unit (unit is resolved server-side; we do not send model)
• --allow-cmd <prefix> – allowlist a command prefix (still prompts to run)
• --show-permissions – print current permissions/policy summary at startup
• --yes – auto-approve safe file writes (never auto-approves commands)

REI-specific subcommands:

• rei agents:get – fetch agent details (REI REST API)
• rei chat:test [message] – minimal chat completion test
• rei mcp list – list MCP servers from the registry
• rei mcp enable <id> – enable an MCP server in .rei/mcp.json

State & config

Repo-local state lives in .rei/ (see REI.md):

• .rei/config.json – repo overrides
• .rei/sessions/ – session logs (JSONL; retention and redaction supported via SessionStore options)

Debugging

Set REI_DEBUG=1 to print request/response envelopes, retry attempts, and provider errors (payloads redacted, unit noted but not sent):

REI_DEBUG=1 node dist/index.js "hello"

See DEBUGGING.md for more.

Testing

bun test

See TESTING.md for the testing taxonomy and near-term coverage goals.

Embedding / IO purity

REPL IO is injectable (write/info/warn/error/debug/log). This enables embedding in TUIs/webviews and transcript capture without stdout scraping. See EMBEDDING_USE_CASES.md for concrete scenarios.

Contributing / repo guide

See REI.md for architecture, conventions, and safety/UX expectations. For migration notes, see BUN_MIGRATION.md.

Roadmap & parity

PRD.md contains the parity charter and current workstreams. Short-term roadmap items are tracked there.

Beta status (current)

Shipped in beta:

• Provider hardening (capabilities, retry/backoff, structured debug logging)
• Session persistence/resume (.rei/sessions/*.jsonl)
• Deny-by-default command policy with allowlist + approval prompts
• MCP discovery and interactive trust flow (/mcp discover, /mcp inspect, /mcp connect)
• REPL IO purity (injectable IO channels used across REPL/slash flows)

Deferred to post-beta:

• Streaming model output
• Checkpoint/hook/plugin parity beyond command plugins
• Full parity for all Claude Code subcommands/flags