What is Monad?
Monad is an open-source agent team runtime from Monadix. One long-lived local daemon owns the identity, capabilities, permissions, memory, sessions, collaboration state, approvals, and audit history of your agents, and keeps them when clients close, when the daemon restarts, and when you swap the runtime backing a team member. It installs as one binary and runs as one process on the machine you already use. See concepts.What does daemon-first mean?
Daemon-first means the durable state lives in a background process rather than in an application window. The Web UI, CLI, TUI, editor bridges, APIs, and messaging channels are clients that render and control that state; none of them owns it, and closing one does not end the work. See runtime, configuration, and security.How is Monad different from an agent framework such as LangGraph or CrewAI?
A framework is a library inside your process that expresses agent logic. Monad is a process that outlives your script: it owns credentials, gates tool calls behind human approval, confines child processes in an OS sandbox, and keeps an audit trail. Framework-built agents can join a Monad team over the Agent Client Protocol. See the full comparison.Can Claude Code, Codex, or Gemini CLI join a Monad team?
Yes. A team member can be backed by a third-party agent provider such as Codex, Claude Code, Gemini CLI, or Qwen Code, by an agent reached over the Agent Client Protocol, by a peer daemon you own, or by Monad Agent Runtime, the first-party runtime Monad bundles. A member backed by a third-party runtime keeps that runtime’s own execution and permission model; Monad records its activity and proxies its approval prompts where the provider exposes them. See use third-party agents.Does Monad send my code or data to the cloud?
Monad sends no telemetry, analytics, crash reports, or usage pings, and stores its state on your machine. Requests to the model providers you configure still leave the machine, because that is where inference happens. Read privacy for exactly what is stored and what crosses the network.What operating systems does Monad run on?
macOS, Linux, and Windows. The daemon, CLI, and Web UI target all three, and continuous integration runs the full suite on each. See installation.Do I need Kubernetes, a message broker, or a database server?
No. Monad is a runtime, not a platform you deploy. It is a single binary and one process on your own machine, with no cluster, message server, gateway, or object store.Is Monad free and open source?
Yes. Monad is released under the MIT license by Monadix Labs, Inc. The source is at github.com/Monadix-AI/monad.How does Monad keep an agent from damaging my machine?
Under Monad Agent Runtime, tool calls are gated by an approval gate, child processes are confined by an OS-level sandbox (Seatbelt on macOS, Landlock and seccomp or bwrap on Linux, AppContainer on Windows), network egress is filtered, and every decision is recorded. See sandbox backends and the runtime security model.Which model providers does Monad support?
Monad ships 24 built-in provider types. Eight use a dedicated SDK — Anthropic, OpenAI, OpenRouter, Vercel AI Gateway, Google Gemini, Mistral, Amazon Bedrock, and Azure OpenAI — and 16 are presets over the bundled OpenAI-compatible adapter, covering Groq, xAI, DeepSeek, Together, Fireworks, Cerebras, Perplexity, Moonshot, Z.AI, MiniMax, NVIDIA, Novita, Ollama, Hugging Face, Cloudflare AI Gateway, and any other OpenAI-compatible endpoint. Custom providers are added as Atom Packs without patching the daemon. See model providers.Can Monad talk to Model Context Protocol servers?
Yes. MCP servers are one of the ways a Monad agent gains capability, alongside skills, Atom Packs, hooks, channels, and agent adapters. See MCP.Can I reach my agents from a phone or a chat app?
Yes. Monad ships 17 channel adapters — including Telegram, Discord, Slack, WhatsApp, Signal, IRC, Microsoft Teams, Google Chat, Feishu, WeCom, LINE, iMessage, email, and a generic webhook — so a session can be driven from a chat conversation and continued in the browser, the CLI, or the terminal UI. See channels.Can I reach the daemon from another machine?
Only if you turn it on. Monad binds local interfaces by default. Remote access is explicit opt-in, requires a bearer token for every non-loopback request, and should sit behind TLS — a reverse proxy, SSH tunnel, or VPN. See remote access.When should I not use Monad?
Monad is the wrong tool in four cases. It is not a hosting platform: if you are shipping an agent-powered product to your customers, you want a serving platform with autoscaling and multi-tenancy, not a single-machine daemon. It is not an authoring framework: if what you need is a way to express agent logic as a graph inside your own application, use a framework and, if it helps later, attach it to Monad over ACP. It does not do cross-owner collaboration: peer federation delegates between daemons you own, and work spanning different owners with independent trust and billing belongs to Monadix. And it does not schedule work across a fleet — there is no distributed scheduler and no universal task schema, so a cluster-scale workload is out of scope. Two more limits worth knowing before you commit. Monad Mesh is alpha and Monad Agent Runtime is experimental, so the API can change between releases. Andnet: 'filtered' egress control
is enforced at the application layer on every platform: a child process that opens a raw
socket bypasses the domain allowlist everywhere except macOS. See
sandbox backends for what each platform actually enforces.