All your AI, one inbox.

Bring the models you already use into Agent Bond, whether that's ChatGPT, Claude, Gemini, Grok, or an agent you run yourself, and message, call, and work with them right next to the people you already talk to.

Message any modelChat with the models you already use, all in one thread list.

Call it tooVoice and video calls with agents, in the same app you use for people.

No migration, no importsYour provider chats stay where they are. Agent Bond adds a home for them.

For everyone · no setup

Try the whole app before you connect anything.

Add a Demo agent and you can message it, call it, and invite it into a group. No keys, no accounts, nothing technical. When you're ready for your real models, the demo agent is a two-second swap away.

Demo, try it now

No keys, no setup. Replies are simulated so you can explore messaging, calls, and groups right away.

1Choose Demo when adding an agent.
2Message, call, and invite it to a group.
3Switch the agent to a real provider whenever you're ready.

Best for everyone, about a minute, zero technical steps.

Bring your own AI · for people who already use APIs

Connect the models you already pay for.

If you have API access to a provider, connecting takes a few minutes: paste a key once, and the model shows up in your inbox. Your key stays on your device and is only ever used to talk to the provider you chose.

API key, the standard way

Paste a key from the AI provider. Agent Bond talks to the provider's official API directly, and the key never leaves your device.

1Create an API key on the provider's site (links below).
2Copy the full key.
3In Agent Bond, tap Add agent → Connection → API key and paste it.
4Pick the model, then tap Test connection to make sure it works.

Best for: people who already have API access and want their real models in one inbox.

MCP server, for self-hosted agents

Point Agent Bond at a Model Context Protocol server for agents you run yourself and custom tooling.

1Get your agent's MCP server URL (from your agent platform, or a tool you run).
2Tap Add agent → Connection → MCP server.
3Paste the server URL, and the bearer token if the server requires one.
4Leave the tool name as "chat" unless the server uses a different name.

Best for advanced setups, agents exposed as MCP tools, and self-hosted systems you control.

One honest thing: for services like ChatGPT, Claude, and Gemini there's no official, secure way for a third-party app to sign in to your account or import your existing chat history, so Agent Bond doesn't fake that. It connects through each provider's API key instead. Your provider chats and your Agent Bond chats stay separate, and keys are stored on your device, never sent anywhere else.

Supported providers

Anthropic (Claude), API key from console.anthropic.com.

OpenAI (ChatGPT/GPT), API key from platform.openai.com.

Google (Gemini), API key from aistudio.google.com.

xAI Grok, API key from console.x.ai.

Meta (Llama), API key + base URL.

Mistral, API key from console.mistral.ai.

OpenClaw, any OpenAI-compatible endpoint (OpenRouter, Together, Ollama, …).

Hermes, a custom OpenAI-compatible gateway.

Custom HTTP, your own endpoint.

Local / self-hosted, Ollama, LM Studio, llama.cpp, vLLM via the OpenClaw path.

Atomic Agent (and Sama), via the OpenClaw path while they're on the roadmap.

For builders & self-hosters

Connect your own assistant.

Built a personal assistant on OpenClaw, Hermes, an MCP server, a custom endpoint, or your own hardware? Give your agent builder one prompt and it hands back exactly the values Agent Bond needs. Paste them into the app, test the connection, and your assistant becomes a contact you can message, call, and add to groups on every device you sign in on.

Step 1, give this to your agent builder

I'm connecting my AI assistant to the Agent Bond app and need it to be reachable through a standard interface. Please expose it (or tell me where it already is) using one of these two options:

OPTION A: OpenAI-compatible HTTP endpoint
Give me these four values:
1. Base URL: the endpoint that accepts POST /chat/completions with OpenAI-style { model, messages } JSON, e.g. https://my-assistant.example.com/v1
2. API key or bearer token: the auth my assistant expects (or "none" if unauthenticated)
3. Model name: exactly as my assistant's API lists it (or "any" if it accepts any)
4. Optional extra: any custom headers my assistant requires

OPTION B: MCP server
Give me these three values:
1. MCP server URL: the HTTP endpoint that speaks the Model Context Protocol JSON-RPC transport
2. Bearer token: if the server requires one (or "none")
3. Chat tool name: the MCP tool to call for chat (defaults to "chat")

Reply as a plain key: value block only, no prose, so I can paste it directly:
  base_url: ...
  api_key: ... (or none)
  model: ...
  mcp_url: ... (only if using Option B)
  mcp_token: ... (or none)
  mcp_tool: chat

Step 2, paste the values and test before you save

1 Get the values Your builder replies with base URL, key, and model, or an MCP URL and token.

2 Paste them in In Agent Bond, tap Add agent → Connection. Choose API key or MCP server, then paste the key: value block from the reply.

3 Test the connection The app round-trips a real message to your endpoint and shows the reply. A broken key or URL is caught here, never after you save.

4 Save and talk Your assistant lands in your inbox with a name and voice. Message it, call it, and bring it into groups.

Every connection is tested for real. Agent Bond sends a live message and shows the actual reply.

Bring your AI when you're ready.

Start free. Message, call, and group-chat with a Demo agent before you connect a single key. Download Agent Bond and see it for yourself.

Free includes 200 demo messages a month. Upgrade now to make it unlimited.