Your own team of AI agents
Give each one its own skills and tools, and let them share what they learn.
Prefer the terminal? Try the openheim CLI
One app, many agents.
Set each agent up for one kind of work. It keeps its own settings, chats and memory.
An agent for each job
A separate agent for each role, with its own chats and memory.
Its own instructions
A system prompt, a default model and the skills it always loads.
Only the tools it needs
Its own MCP servers and subagents, and you decide what runs without asking.
Folders to work in
Point it at folders on your machine. Chats stay grouped by folder.
Shared memory
What one agent learns, the others can find. Works with any model.
Local or remote
Run agents in the app, or on your own openheim server.
Watch it work. Step in when it matters.
Everything openheim can do, without a config file in sight.
- 01
See every step
Thinking, text and tool calls stream in as separate blocks, in the order they happen.
- 02
Approve what runs
Each tool call waits for you: allow once, allow always, or reject. Calls made by subagents ask too.
- 03
Code or Architect
Code mode has every tool. Architect mode only gets read-only tools, so the agent plans but never edits or runs anything. Set the reasoning effort per chat too.
- 04
Get pinged
When a chat you don't have open finishes, fails or needs approval, you get a notification, even from another agent.
Switch models mid-chat
Pick another model and the next message uses it.
Redirect a running turn
Type while it works to stop the turn and start over.
MCP servers
Connect any MCP server. Its tools show up right away.
Skills
Markdown instructions you turn on per chat or always.
Subagents
Hand a task to another agent, get back only the answer.
Files
Browse the agent's folder in tabs that refresh live.
Paste images
Drop a screenshot or photo into the prompt box.
Built for long chats
A ring shows how full the context window is.
Bring your own model.
Add your API key once and every agent can use it. Each agent picks its own default, and you can switch in the middle of a chat.
- Anthropic
- OpenAI
- Google Gemini
- xAI
- DeepSeek
- Kimi
- Z.ai
- Ollama
Plus anything that speaks the OpenAI API, like OpenRouter, vLLM or LM Studio. Add it as a custom provider with its base URL.
Private by default.
You choose where your conversations go. Nobody else sees them.
Read the privacy policyNo ads, no tracking
Only anonymous usage statistics, never your chats or files. Turn them off any time in Settings.
Your data stays on your device
Conversations, settings and API keys are stored locally. Messages go only to the provider or server you choose.
Nothing runs without asking
Tool calls wait for your approval unless you've chosen to always allow them.
Files stay in their folder
Each chat works in one folder, the agent's own workspace or one you added. File tools can't read or write outside it.
Common questions
More on the support page.
Is Openheim Studio free? +
Yes, the app is free. You pay your model provider for what you use, or nothing at all with a local model.
What is an agent in Openheim Studio? +
An agent is an AI assistant set up for one kind of work. Each one has its own system prompt, skills, MCP servers, subagents, folders, chats and memory, and picks its own default model. Make as many as you need and switch between them at the top of the sidebar.
Can my agents share what they know? +
Yes, if you want them to. Every agent keeps its own memory. Turn on shared memory for two or more agents and they also read and write one memory they all share. It works with any model.
Is it available on iPhone and Android? +
The iOS and Android apps are in review on the App Store and Google Play. The store links will appear on this page once they are live. The desktop app is available now for macOS, Windows and Linux.
Which AI models can I use? +
Anthropic, OpenAI, Google Gemini, xAI, DeepSeek, Kimi, Z.ai and Ollama work out of the box. Anything that speaks the OpenAI API, such as OpenRouter, vLLM or LM Studio, can be added as a custom provider with its base URL. You set up a provider once and every agent can use it, and you can switch models in the middle of a conversation.
Can I run it with a local model? +
Yes. Point a provider at Ollama or any other OpenAI-compatible server running on your machine. No API key is needed, and your conversations never leave your computer.
Does it work with the openheim CLI? +
Yes. Each agent lives in a folder laid out like the CLI's own, so you can run openheim --config with the agent's config.toml to use the same agent in a terminal. Chats from the CLI show up in the app as they happen. You can also connect the app to an openheim server running elsewhere.
Can the agent run shell commands? +
On desktop, yes, once you turn the shell on for the agent or for one of its folders. It asks before each command unless you choose to always allow it. iOS and Android don't let apps run other programs, so on mobile you connect to an openheim server on your computer to use its shell.
Can I connect MCP servers? +
Yes. Add an MCP server to an agent over stdio or Streamable HTTP, and its tools become available to that agent. On mobile, only HTTP-based MCP servers are supported.
