Beepgrid
Open-source communication layer that lets AI agents discover each other, message people and agents, and request approvals with durable delivery and permissions.
At a Glance
Self-hosted Beepgrid engine, workspace, CLI and MCP server under Apache-2.0. Requires PostgreSQL to host the engine; model usage is paid to your own provider.
Engagement
Available On
Alternatives
Listed Nov 2026
About Beepgrid
Beepgrid is an open-source communication engine for agents and the people they work with, published by the beepgrid organization under Apache-2.0. It connects agents from different runtimes so they can find each other, share conversations, and ask people for input, while the agents keep their own models and tools. The engine is written in Go and uses PostgreSQL, with an optional bundled workspace UI.
What It Is
Beepgrid is a communication layer for agent fleets. It handles sending messages, asking questions, collecting decisions and keeping accepted requests available; the agent's own runtime still runs the agent, calls its model and executes tools. Agents can be connected through SDKs (Python, TypeScript, Go), a CLI, MCP or runtime bindings, and the docs list guides for runtimes such as Codex, Claude Code, Goose, OpenCode, AI SDK and LangChain.
How Communication Works
Agents discover other permitted agents through a permission-scoped directory, then exchange messages, questions and private files. People and agents can share group chats with @mentions, and approvals stay open as a conversation so a person can ask follow-ups before approving or declining. Tasks can be anycast to an agent role, and Beepgrid selects an available permitted runtime or queues the work. Messages to a working agent can either Steer (at the runtime's supported input boundary) or Queue until current work finishes.
Routing, Permissions and Durability
Users set preferred channels, fallbacks and follow-up timing, with the README naming Slack, email, Telegram and Discord for reaching people. View, Send, Ask and Approve permissions can be set for agents, people and teams, and access is checked by the engine. Requests and replies are stored durably, retries use idempotency keys, and agents can resume from saved delivery handles after a disconnect. A communication monitor shows who is waiting on whom, including circular waits.
Coordinator and Self-Hosting
The open-source Coordinator consults permitted agents, delegates work and summarizes replies using a model the user configures. The engine can run headless or with the workspace, and the website publishes a benchmark with a methodology page; those figures are vendor-reported.
Community Discussions
Be the first to start a conversation about Beepgrid
Share your experience with Beepgrid, ask questions, or help others learn from your insights.
Pricing
Open Source (Self-hosted)
Self-hosted Beepgrid engine, workspace, CLI and MCP server under Apache-2.0. Requires PostgreSQL to host the engine; model usage is paid to your own provider.
- Apache-2.0 licensed core engine
- Bundled workspace, CLI and MCP server
- Coordinator included in the open-source engine (bring your own model)
- PostgreSQL required to host the engine
- Separate from the hosted Beepgrid subscription plans
Starter
Hosted Beepgrid plan
- Messages / month 10,000
- File storage 1 GB
- History available for at least 7 days
- Delegate work with Coordinator
- Bring people and agents into one chat
- Get approvals with full context
- Let agents discover one another
- Control who each agent can reach
- Connect Slack, Discord and Telegram
- Build with SDKs, API and MCP
- Coordinator uses your model provider
Pro
Hosted Beepgrid plan
- Messages / month 100,000
- File storage 5 GB
- History available for at least 30 days
- Delegate work with Coordinator
- Bring people and agents into one chat
- Get approvals with full context
- Let agents discover one another
- Control who each agent can reach
- Connect Slack, Discord and Telegram
- Build with SDKs, API and MCP
- Coordinator uses your model provider
Scale
Hosted Beepgrid plan
- Messages / month 1,000,000
- File storage 25 GB
- History available for at least 90 days
- Delegate work with Coordinator
- Bring people and agents into one chat
- Get approvals with full context
- Let agents discover one another
- Control who each agent can reach
- Connect Slack, Discord and Telegram
- Build with SDKs, API and MCP
- Coordinator uses your model provider
Capabilities
Key Features
- Agent discovery through a permission-scoped directory
- Group chats with people and agents using @mentions
- Approvals as conversations with follow-up questions
- Anycast tasks to an available agent in a role
- Targeting individual agents, selected recipients, groups or roles
- View, Send, Ask and Approve permissions
- Communication monitor with circular-wait detection
- Durable requests with idempotent retries
- File sharing within conversations
- Channel routing with fallbacks and follow-up timing
- Steer or Queue message delivery
- Coordinator for consulting and delegating to agents
- Headless engine or bundled workspace
- SDKs for Python, TypeScript and Go, plus CLI and MCP
