Cindy

Open-source agent platform that orchestrates Claude Code, Codex, and other coding agents on a shared workspace

Freemium· Free (BYO API keys) / Plus from $20/mo (managed model gateway across GPT, Claude, Grok, Gemini, DeepSeek, Kimi, GLM, Qwen) / Team (coming soon, SSO + credit controls) / Enterprise (custom, self-hosted gateway + audit logs)AgentsModel-agnostic: Claude (via Claude Code), GPT (via Codex), Grok, Gemini, DeepSeek, Kimi, GLM, Qwen, plus local models via OpenAI-compatible endpoints and Ollama

Visit website →

Best for

Individual developers and small engineering teams who already use Claude Code or Codex, want to run agents in parallel, and refuse to lock their workflow into a single vendor's ecosystem.

Skip if

Non-technical users who want a hosted chatbot, or enterprises that need mature SSO and governance today rather than waiting for the Team tier.

Cindy is an open-source (Apache-2.0) agent runtime that sits above individual coding assistants and lets you drive them from one workspace. Instead of picking a single vendor lock-in stack, you attach one or more 'harnesses' — currently Claude Code and OpenAI Codex, with more planned — and route each phase of a task (planning, editing, review) to whichever harness or model handles it best, without losing memory, tools, or working files when you switch. The platform is local-first: repositories, credentials, and intermediate state stay on your machine by default, and every agent action lands in an isolated workspace that you approve line-by-line before it touches your real branch. Multi-agent parallelism is a first-class primitive, so you can fan out plan/execute/review across models and reconcile the results. Cindy also ships integrations with the surfaces engineering teams actually work in — Slack, Feishu, GitHub, Jira, and remote SSH hosts — plus support for MCP tools, custom skills, and voice interfaces, so a task kicked off from a Slack thread can end as a reviewed PR. Typical workflows include long-running refactors that hand off between a strong planner and a cheap executor, autonomous bug triage against a live repo, and running the same task through two agents in parallel to compare diffs. The Free tier is fully functional if you bring your own provider keys; the Plus subscription bundles a managed gateway across all major frontier and open-weight models so you don't have to juggle billing accounts.

Editor's take

Cindy is the clearest attempt yet at a vendor-neutral shell around the coding-agent wars — you keep Claude Code and Codex, add whatever comes next, and don't rewrite your workflow each time. Local-first plus line-by-line approvals means it fits how careful engineers actually work. The gap is governance: Plus is a solo tool, Enterprise is a sales call, and there's nothing in between yet.

— The AI Tool Bible editorial team

Pros

✅ Apache-2.0 licensed with a public GitHub repo — self-host or fork without vendor risk
✅ Harness abstraction lets you swap Claude Code, Codex, and future agents mid-task without losing workspace state
✅ Local-first: code, keys, and diffs stay on your machine unless you opt into the managed gateway
✅ Isolated workspaces with line-by-line diff approval before any change hits your real branch
✅ Native parallel multi-agent orchestration for plan/execute/review pipelines
✅ First-party integrations with Slack, Feishu, GitHub, Jira, and SSH so agents plug into real team workflows
✅ Managed Plus tier gives one bill covering GPT, Claude, Grok, Gemini, DeepSeek, Kimi, GLM, and Qwen

Cons

⚠️ No public developer API to embed Cindy inside other products — MCP is inbound only
⚠️ Value depends on having Claude Code or Codex credentials already; not a standalone coding model
⚠️ Team tier with SSO and credit governance is still 'coming soon', so mid-size orgs face a gap between Plus and Enterprise
⚠️ Multi-harness orchestration adds cognitive overhead versus using one agent directly
⚠️ Ecosystem is young — fewer community skills and playbooks than Cursor, Cline, or vanilla Claude Code

Use cases

Multi-agent code refactorsParallel plan / execute / review pipelinesAutonomous bug triage on a live repoSlack-triggered coding tasksJira ticket to pull request automationCross-model diff comparisonLocal-first agent execution with human approvalRemote SSH server administration via agentMCP tool orchestration across harnesses