Your AI Dev Team.
In Your Terminal.

Reeve Code is a CLI-first AI coding agent with multi-agent orchestration, persistent memory, and real-time cost tracking. Not another copilot — an entire engineering team in your terminal.

Free forever • BYOK supported • 100+ managed models

Reeve Code — Terminal
Agent Chat
Sessions
coordinator

Analyzing your request. Spawning 2 sub-agents for parallel refactor...

code-worker-1

Refactored auth.service.ts — extracted token refresh into composable. 42 lines reduced.

test-runner

All 24 tests passing. Coverage: 94%. Ready for review.

Ask your agents anything...
Reeve Connected3 agents active
$0.42 this sessionClaude Opus 4
Core Features

Everything you need. One command.

Multi-agent orchestration, persistent memory, and cost tracking — all from your terminal.

CLI-First Design

Runs in your terminal. No IDE required. Works everywhere — SSH, tmux, CI/CD, your laptop.

Multi-Agent Orchestration

Coordinator delegates to workers. Parallel execution, fresh context per task, automatic retries.

Pipeline

From task description to shipped code. PRD → design review → implementation → testing → audit. All automated.

Live Session Monitoring

Watch agents work in real-time. See tool calls, token spend, and progress as it happens.

Cost Tracking

Know exactly what you're spending. Per-session costs, model breakdowns, budget alerts. Full attribution across agents and teams.

Persistent Memory

Agents remember across sessions. Daily logs, long-term memory, vector search — context that compounds.

Architecture

Three layers. CEO / VP / IC.

Instead of one agent doing everything (and drowning in context), work flows through three layers — modeled after how companies actually operate.

Layer 1

Coordinator

CEO

Wide + shallow context. Routes requests, directs work, verifies results. Talks to the user. Never touches code or files directly.

Summaries from many projectsUser conversationRouting decisionsHigh-level status
Layer 2

Project A Mgr

VP — Owns Frontend app. Spawns workers, runs pipelines, reports up.

~15m
Layer 2

Project B Mgr

VP — Owns API service. Spawns workers, runs pipelines, reports up.

~15m
Layer 2

Project C Mgr

VP — Owns Mobile app. Spawns workers, runs pipelines, reports up.

~15m
Layer 3

Worker

IC — 1 task
~1h
Layer 3

Worker

IC — 1 task
~1h
Layer 3

Worker

IC — 1 task
~1h
Layer 3

Worker

IC — 1 task
~1h
Layer 3

Worker

IC — 1 task
~1h
Layer 3

Worker

IC — 1 task
~1h

The Key Insight: Same Capacity, Different Content

All three layers use ~50-60% of their context window. The difference is what fills it. A fresh worker gets a clean 200k-token window filled entirely with the files and test output it needs — no conversation history, no other projects, no routing decisions polluting its focus.

Worker ~1hr
Manager ~15min
Coordinator ~4min

1:4 ratio at every layer boundary — if the coordinator spends as much time as the worker, the architecture is broken.

Context Fill Comparison

Coordinator
CEO
~55%
Many project summariesUser conversationsRouting decisions
Depth:Shallow
Span:Wide
Manager
VP
~58%
One domain deepRepo architectureRecent changes & errors
Depth:Medium
Span:Medium
Worker
IC
~55%
One task deepSpecific files & testsError traces
Depth:Deep
Span:Narrow
Pipeline

From task to shipped code.

The pipeline turns a task description into working, tested, audited software — with truly parallel multi-agent execution.

Task

PM

PRD

Architect

Design Review

parallel
ArchitectSecurityQA

Implementation

parallel
BackendFrontend...

Testing

QA Agent

Audit

Security
8 concurrent agents
Semaphore-bounded parallel execution
Retry with backoff
Failed tasks retry up to 2x before blocking
Smart model routing
Opus → architecture, Sonnet → code, Haiku → perf
Real-time progress
.pipeline/progress.json for live monitoring
Dependency graph
Tasks run when their deps are satisfied
Deadlock detection
Graceful failure when no tasks can proceed
Tools & Skills

Every tool. One runtime.

Agents have access to a complete toolkit — from file editing to browser automation to companion device control. Skills are pluggable and follow the AgentSkills-compatible format.

File System

Read, write, edit files

Shell Execution

Run commands, scripts

Web Search & Fetch

Brave Search, URL extraction

Browser Automation

Full browser control

Node Control

iOS / Android / macOS companions

Cron Scheduling

Heartbeats & timed tasks

Message Routing

Send across all channels

Memory System

Vector search, daily logs, long-term

Sub-Agent Spawning

Parallel delegation, fresh context

Gateway

One control plane. Every surface.

A single Gateway daemon coordinates agent sessions, channel connections, tool execution, and companion nodes — everything flows through one process.

WhatsApp
Telegram
Slack
Discord
Signal
iMessage
Teams
WebChat

Gateway

Control Plane — ws://127.0.0.1:18789

Channel routingSession managementAuth & pairingWebSocket API

Agents

Isolated workspaces, memory, sessions. RPC over WebSocket.

Skills & Tools

Pluggable capabilities via AgentSkills-compatible format.

Nodes

iOS, Android, macOS companions. Camera, voice, Canvas.

Integrations

8 channels. All native.

Not just APIs — full bidirectional chat support on every platform. Channels run simultaneously and route per chat.

WhatsApp

Baileys (Web protocol)
DMs, groups, media, reactions

Telegram

grammY (Bot API)
DMs, groups, topics, voice

Discord

discord.js
DMs, servers, threads

Slack

Bolt SDK
DMs, channels, threads

iMessage

imsg CLI / BlueBubbles
macOS native

Teams

Bot Framework
DMs, channels

Signal

signal-cli
DMs, groups

WebChat

Built-in
Local or remote UI
Context Management

Long sessions. Zero degradation.

Most frameworks degrade as context fills. Reeve uses cache-TTL pruning, auto-compaction with memory flush, and per-layer budgets to keep agents sharp indefinitely.

Reeve — Stays Fresh

Turn 1
12%
Turn 50
45%
Turn 100
52%
Turn 200
48%
Cache-TTL pruning — trims old tool results, tuned for Anthropic cache hits
Auto-compaction — summarizes old conversation, flushes memory to disk first
Per-layer budgets — Coordinator <20%, Manager <70%, Worker 100%

Other Frameworks — Context Bloat

Turn 1
12%
Turn 50
55%
Turn 100
82%
Turn 200
97%
No pruning — all tool output stays forever
No compaction — context fills until OOM or truncation
Single agent holds everything — quality degrades with length
Pricing

Free for individuals. Pro for teams.

The full CLI is free — all features, no gates. Go Pro when your team needs shared agents and memory.

Save 20% ✨

Free

$0

Full CLI. All features. No catch.

  • Full Reeve CLI
  • Unlimited local agents
  • All channels
  • BYOK — any API key
Most Popular

Pro

$20/mo per user

Shared agents, memory, and skills for teams.

  • $20/mo in Base Credits included
  • All first-party apps
  • Shared memory & skills
  • Up to 25 users

Cloud

$40/mo per user

Always-on cloud gateway. Unlimited scale.

  • $20/mo in Base Credits included
  • Cloud gateway (24/7)
  • SSO / SAML
  • Unlimited users
Works with Claude Code Max subscriptions

View full pricing details

30 Days Free

Ready to build with
an AI team?

Install Reeve Code. Start building with multi-agent orchestration.
Free for individuals. Pro for teams.

// No credit card required. No lock-in. Just better code.