Build agents once.Run them from your assistant over MCP.
XonaMind is the platform where your agents live: prompt, model, tools, memory, skills and knowledge bases. Run them by API, widget or MCP. Bring your own provider keys or use our credits (real cost + 10%).
No card to try it. Works from Claude, ChatGPT, Cursor or your own code.
support-bot
Support agent
- Model
- GPT-5 · OpenRouter
- Tools
- HTTP · Web search
- Memory
- Conversations and preferences
- Knowledge
- Help center · 128 docs
- Skills
- Refunds · Triage
Runs through MCP, API or widget with the same agent.
How it works
From idea to production in three steps
An agent is a prompt, a model, tools and memory. You build it; XonaMind runs it.
- 1
Create the agent
Pick a model, write the prompt and add memory, skills and knowledge bases.
- 2
Connect tools and knowledge
Add HTTP or OpenAPI tools, webhooks, schedules and OAuth connections for your end users.
- 3
Call it from where you work
API, OpenAI-compatible endpoint, streaming widget or MCP from your assistant.
Everything included
The infrastructure behind every agent
You do not build queues, retries, traces or provider billing. XonaMind already does.
Durable runs
Runs with retries, state and history. If something fails, you see it and requeue it.
Webhooks
Tell your app when a run finishes, fails or needs approval.
Schedules
Run agents on a cron without maintaining your own workers.
Evals
Measure answers against a set of cases before changing the prompt or model.
Traces
Every agent step, with tools, latency and tokens.
Metrics and alerts
Usage, cost, errors and latency per agent, with alerts when something goes out of range.
Your keys or our credits
Use your own keys (Anthropic, OpenAI, OpenRouter, fal, Venice, DeepSeek…) and the provider bills you directly, or use XonaMind models with credits at real cost + 10%.
HTTP and OpenAPI tools
Connect any API with an OpenAPI schema or your own HTTP tool.
Knowledge and memory
Documents with vector search, plus per-user and per-conversation memory.
Credentials vault
Store your provider keys and tool secrets encrypted, without exposing them to your users.
End-user OAuth
Your customers connect their accounts without you seeing or storing their tokens.
Also: every module on every plan, scoped personal tokens, per-workspace roles and audit logs.
Shared memory
Shared memory for your team and your agents
A brain that your whole team's assistants (Claude Code, Cursor, Codex) read and write through MCP, alongside your agents.
Memory bases
Split knowledge by team, product or topic, and choose which tools and agents read or write in each one.
Review before it counts
What your tools learn enters as a draft and does not affect answers until someone approves it.
Graph of related knowledge
Memories link to each other, so recall brings back what is related, not just what looks similar.
Overnight dreaming
Overnight XonaMind spots duplicates and contradictions, links memories and organizes knowledge into areas.
Works without building agents: connect your tool through MCP or the API and you are done. Included in every plan.
Native MCP
Your whole workspace, from your assistant
XonaMind exposes the same product over MCP. Create agents, edit prompts, connect tools, read traces and manage keys without opening the dashboard.
Open your client
Claude, ChatGPT, Cursor or any MCP-compatible client.
Point it at /mcp
Use your API URL and a personal token for your workspace.
Ask in plain language
"Create an agent that triages tickets", "show me today's traces", "add a webhook".
{
"mcpServers": {
"xonamind": {
"url": "https://api.xonamind.app/mcp",
"headers": {
"Authorization": "Bearer xm_pat_your_token"
}
}
}
}Replace the example token with a personal token from your workspace. The token is never shared with the model.
Then just ask
"Create a support agent with GPT-5, web search and my knowledge base. Publish it over the API and notify me by webhook."
Integrations
Call your agents from anywhere
The same API for your backend, your frontend, your assistant or your customers' chat.
REST API
Runs, streaming, history and webhooks from your backend.
curl -X POST "https://api.xonamind.app/v1/agents/support-bot/runs" \
-H "Authorization: Bearer xm_pat_your_token" \
-H "Content-Type: application/json" \
-d '{"input": "Summarize today's tickets"}'OpenAI-compatible
Change the base URL and use the OpenAI SDK you already have.
curl "https://api.xonamind.app/v1/openai/chat/completions" \
-H "Authorization: Bearer xm_pat_your_token" \
-H "Content-Type: application/json" \
-d '{
"model": "xonamind/support-bot",
"messages": [{ "role": "user", "content": "Summarize today's tickets" }]
}'Embeddable widget
A streaming chat for your site, with no UI to build.
<script
src="https://api.xonamind.app/widget.js"
data-agent="support-bot"
async
></script>SDK
A typed client to ship agents in minutes.
import { XonaMind } from "@xonamind/sdk"
const xm = new XonaMind({ apiKey: process.env.XONAMIND_API_KEY })
const run = await xm.agents.run("support-bot", {
input: "Summarize today's tickets",
})Pricing
Start small, scale when you need it
Monthly subscription with every module included; plans only change the limits. No hidden fees.
- Limits per plan
- Trial credits on the free plan
- Change or cancel anytime
FAQ
What to know before you start
Agents, models, costs and MCP, without the fluff.
What is an agent in XonaMind?+
It is a reusable configuration: prompt, model, tools, memory, skills and knowledge bases. The same definition is called by API, the OpenAI-compatible endpoint, the widget or MCP.
Can I use my own models?+
Yes. Bring your provider keys or use XonaMind, which routes across OpenRouter and direct providers such as OpenAI, DeepSeek, Alibaba, Venice and fal. You can pin a model per agent or let routing pick the cheapest option.
How does billing work?+
You pay a monthly subscription for the platform: every module is on every plan and only the limits change. Model usage is separate: with your own keys the provider bills you; with XonaMind models you use prepaid credits at real cost + 10%. You see the cost in the traces.
Do I need my own API keys?+
No. You can bring your keys (Anthropic, OpenAI, OpenRouter, fal, Venice, DeepSeek…) or use XonaMind models with prepaid credits. You can mix both: a workspace can use its keys for some models and credits for others. The free plan includes a few cents of trial credits so you can start without a key.
How much does it cost to use XonaMind models?+
You pay with prepaid credits at the provider's real cost plus a declared margin, currently 10%. There is no hidden markup. If you bring your own keys, the provider bills you directly and XonaMind charges nothing for those calls.
Are my keys safe?+
They are stored encrypted, never shown again after you save them, and you can remove them at any time.
What happens if a provider fails?+
Runs are durable: they have state, retries and history. If a model fails, the run is recorded and you can retry it; webhooks notify your app.
Do I need the dashboard?+
No. Everything you can do on screen can be done over MCP: agents, tools, credentials, knowledge, webhooks, evals, keys and members. The dashboard is there for those who prefer it.
Where do my users' tokens go?+
Your end users' OAuth connections are stored in the workspace credentials vault. They authorize once and you never handle their secrets.
Can I use only the memory?+
Yes. Connect your tool (Claude Code, Cursor, Codex) through MCP, or your own app through the API, and use XonaMind purely as a shared memory, without creating agents. It is included in every plan.
Your next agent starts here
Create a workspace, build your first agent and call it over MCP the same afternoon.