Start
Providers and keys
OpenIntern runs on your own account with an AI provider. You add the provider's API key in the pane, pick a model, and the provider bills you directly for what your tasks use.
Supported providers
Current OpenAI and Anthropic models are the ones we test. The others work through the same setup; how well a task goes depends on the model.
| Provider | Store add-in | Local software | Get a key |
|---|---|---|---|
| OpenAI | Yes | Yes | platform.openai.com |
| Anthropic | Yes | Yes | console.anthropic.com |
| OpenRouter | Yes | Yes | openrouter.ai |
| Google Gemini | Yes | Yes | aistudio.google.com |
| xAI | Yes | Yes | console.x.ai |
| Groq | Yes | Yes | console.groq.com |
| DeepSeek | Yes | Yes | platform.deepseek.com |
| Mistral | Yes | Yes | console.mistral.ai |
| Together AI | Yes | Yes | api.together.ai |
| Fireworks | Yes | Yes | fireworks.ai |
| Cerebras | Yes | Yes | cloud.cerebras.ai |
| Moonshot (Kimi) | Yes | Yes | platform.moonshot.ai |
| Alibaba Qwen | Yes | Yes | modelstudio.console.alibabacloud.com |
| Azure OpenAI | No | Yes, with your resource's endpoint URL | portal.azure.com |
| Ollama | No | Yes, no key needed | Runs on your computer |
| LM Studio | No | Yes, no key needed | Runs on your computer |
| OpenAI-compatible (custom) | No | Yes, any endpoint you enter | Depends on the endpoint |
The store add-in offers only providers with a fixed public address. Providers whose address you type in (Azure, local runtimes and custom endpoints) would make our service connect to an address a visitor chose, so they are available only in the local software.
Add, test and remove a key
- Choose More (⋯), then Providers.
- Pick the provider from the Provider list. The pane shows the fields that provider needs: a key, with a link to where you get one, and for Azure, local runtimes and custom endpoints, a base URL.
- Paste the key and choose Save. The pane can fetch the provider's list of models so you can choose from it.
- Choose Test connection. OpenIntern sends one minimal request through your key and reports the time it took, or the provider's error.
To take a key out, open the provider under Providers and choose Remove. Removing it from OpenIntern does not cancel it: revoke it in your provider's console as well if you no longer need it.
Where your key is kept
In the store add-in
Your key is kept in the add-in's own storage on this device, the browser storage Office keeps for the add-in. It is not saved inside your documents and not saved on our servers. The pane sends it over HTTPS with each request to app.openintern.ai, and to no other website. Our service takes the key out of the request as soon as it arrives, uses it only to call the provider it belongs to, holds it in memory while that task runs, and never writes it to a database, file or log.
Because the key travels to our service with each request, our service handles it while it works. If you would rather your key never leave your computer except to go to your provider, use the local software.
Removing the add-in or clearing Office's add-in cache deletes the key from this device, so you would need to add it again.
In the local software
Keys you enter in the pane are saved on your computer in ~/.openintern/credentials.json, which only your user account can read. You can instead supply a key in the server's environment, for example OPENAI_API_KEY or ANTHROPIC_API_KEY; a key in the environment takes precedence over a saved one. Keys are never sent back to the pane, which shows only whether a key is set and its last four characters, and they are never written to logs or task journals. Requests go straight from your computer to the provider.
Choose a model
The model picker in the composer lists the models of every provider you have set up. When more than one provider is set up, each model is labelled with its provider. Your choice applies to new tasks, so you can switch models from one request to the next.
Connectors
Exa web search
With a key from Exa, OpenIntern can search the web and read web pages during a task. Add the key under Providers, in the Connectors section, or from the composer's + menu. It is off until you add a key. Search queries can include words from your request, and they go to Exa (api.exa.ai) under Exa's terms. Text on a web page is treated as information, never as instructions.
Local software only
- MCP servers. Add tools from Model Context Protocol servers, local or remote, in
~/.openintern/mcp.json(the same shape Claude Desktop uses). The Providers panel lists them under MCP servers, and remote servers that ask you to sign in show Connect. A tool call to an MCP server counts as a change unless the server marks it read-only. - SEC EDGAR filing lookups. Company facts from SEC filings. The SEC requires a contact name and email address in each request, which you supply.
- Fully local mode. Start the server with
OPENINTERN_NO_EGRESS=1and it refuses every provider that is not on your own computer, so only a local model can be used.
Costs and usage
OpenIntern does not charge for the add-in, the local software or the hosted service. Your provider charges for the tokens your tasks use. More (⋯) > Usage shows model calls, tokens, the share served from the provider's cache and an estimated cost, for today, 7 days, 30 days or all time, with a per-task history. Estimates can differ from your provider's bill; the bill is what counts. OpenIntern uses prompt caching where the provider supports it, which can lower the cost of long tasks.