Introduction

Autonomous agents for every project. Deployed where your team works.

CodeSpar is an open source, multi-agent platform that deploys autonomous AI coding agents directly into the messaging channels your team already uses — WhatsApp, Slack, Telegram, Discord, and CLI. Each project gets its own persistent agent that monitors builds, investigates failures, proposes fixes, and orchestrates deploys, all via @mention commands.

Key Features

• Eight specialized agent types — purpose-built agents for project management, task execution, code review, deployment, incident response, cross-project coordination, sprint planning, and codebase analysis
• Five messaging channels — WhatsApp, Slack, Telegram, Discord, and CLI, all with the same @mention interface
• Graduated autonomy (L0–L5) — fine-grained control over how much independence your agents have, from passive (L0) to fully autonomous (L5)
• Security-first architecture — 10 defense layers including sandboxed Docker execution, secrets isolation, audit logging, and human approval gates for sensitive operations
• Open source (MIT License) — full transparency, community contributions, self-host anywhere

How It Works

1. You talk to your agent in any channel

Just @mention your agent in Slack, WhatsApp, Discord, or Telegram — same syntax everywhere:

@codespar instruct add a health check endpoint at /api/health

2. The Dev Agent reads your code and creates a PR

The agent reads your actual codebase via the GitHub API, sends the relevant files to Claude, and creates a pull request with the changes:

[agent-default] Task completed: task-1773970
  Instruction: add a health check endpoint at /api/health
  Duration: 12.3s

  Changes applied to 1 file(s):
    - app/api/health/route.ts

  PR #1 created: https://github.com/your-org/your-repo/pull/1

3. The Review Agent analyzes the PR

Automatically (or on demand), the Review Agent fetches the PR diff, classifies risk, and posts a detailed code review:

@codespar review PR #1

4. Your team approves from any channel

A developer approves from WhatsApp what was requested in Slack. Cross-channel, token-based approval with quorum:

@codespar approve dp-abc123

5. Everything is audited

Every action — from commands to approvals to deployments — is logged in an immutable, hash-chained audit trail.

Architecture

Under the hood, CodeSpar follows a persistent + ephemeral agent architecture:

┌─────────────────────────────────────────────────┐
│                  Coordinator                     │
│           (cross-project orchestration)          │
└──────────────────────┬──────────────────────────┘
                       │
        ┌──────────────┼──────────────┐
        ▼              ▼              ▼
┌──────────────┐┌──────────────┐┌──────────────┐
│Project Agent ││Project Agent ││Project Agent │
│  (repo-a)    ││  (repo-b)    ││  (repo-c)    │
└──────┬───────┘└──────┬───────┘└──────────────┘
       │               │
       ▼               ▼
  ┌─────────┐    ┌─────────┐
  │Dev/Task │    │ Review  │
  │  Agent  │    │  Agent  │
  └─────────┘    └─────────┘
  (ephemeral)    (ephemeral)

• Project Agent (persistent) — your team's always-on interface. Monitors CI/CD, handles @mention commands, spawns ephemeral agents.
• Dev Agent (ephemeral) — reads your codebase via GitHub API, sends context to Claude, creates branches and PRs with real code changes. Learn more
• Review Agent (ephemeral) — fetches PR diffs, classifies risk (low/medium/high), auto-approves low-risk at L3+. Learn more
• Deploy Agent (ephemeral) — approval workflows with quorum, cross-channel voting, staging and production. Learn more
• Incident Agent (ephemeral) — investigates build failures, correlates with recent changes, proposes fixes. Learn more
• Coordinator (persistent) — cross-project orchestration, cascading deploys, status aggregation. Learn more

The agent also understands natural language in any language. Ask in Portuguese, Spanish, or English — it uses Claude Haiku for intent classification and Claude Sonnet for open-ended questions. Learn more

Autonomy Levels

CodeSpar uses a graduated autonomy model so you control exactly how much independence your agents have:

Safety guardrail: Regardless of autonomy level, agents never auto-execute production deployments, data migrations, security-sensitive changes, or infrastructure modifications. These always require explicit human approval.

Tech Stack

CodeSpar is built with:

• TypeScript — end to end, strict mode
• Fastify — high-performance HTTP server
• Turborepo — monorepo with 13 packages
• Docker — sandboxed agent execution
• MIT License — use it, fork it, contribute to it

Next Steps

Ready to get started? Head to the Quickstart guide to have CodeSpar running in under 5 minutes, or read the full Installation guide for production deployment options.