Skills for the work around the work, a Workbench beside every chat, the model of your choice, and tools that ask before they act.
Type / in the message box, or attach a file and pick a suggestion. Each skill gives your buddy an expert playbook and a strict format, so results come back as cards you can act on.
and 13 more
Break work down and see what could go wrong.
Minutes, summaries and requirements documents.
Diagrams rendered right in the chat.
Updates and emails ready to send.
New Workbench widgets, described in plain words.
The panel beside the chat holds pinned widgets as tabs, with its own layout for each project. Add a built-in widget, or ask a buddy to design a new one.
8 widgets that read and change your real records.
Ask for a tracker or dashboard. The buddy returns a spec, not code; the app checks it, repairs it and draws it with the same parts as the built-in widgets, in light and dark mode. Edit with buddy sends it back to the chat and keeps your rows.
Formula fields, worked out by the app's safe calculator:{value} * {probability} / 100days_until({due})
Agent2Work runs on Ollama out of the box. Switch models from the message box at any time, per conversation.
| Provider | Where it runs | Key |
|---|---|---|
| OllamaRun open models locally on your computer. | Your machine | No key |
| LM StudioServe local models from the LM Studio desktop app. | Your machine | No key |
| OpenRouterOne key for hundreds of hosted models. | Cloud | Your key |
| OpenAIGPT models from OpenAI. | Cloud | Your key |
| AnthropicClaude models from Anthropic. | Cloud | Your key |
| Google GeminiGemini models from Google AI Studio. | Cloud | Your key |
| GroqVery fast inference for open models. | Cloud | Your key |
| MistralMistral's hosted models. | Cloud | Your key |
| DeepSeekDeepSeek chat and reasoning models. | Cloud | Your key |
| xAIGrok models from xAI. | Cloud | Your key |
| Together AIOpen models hosted by Together. | Cloud | Your key |
| Custom endpointAny OpenAI-compatible server: vLLM, llama.cpp, LocalAI, Jan… | Anywhere | Key optional |
Long documents on Ollama switch to a larger-context copy of your model automatically, so nothing gets cut off at 4,000 tokens.
Each buddy has its own allowed tools. Browser integration tokens go straight to GitHub, Google and Microsoft. Finance uses your separate encrypted server connection for prices and Telegram alerts.
Create and update tasks, and remember facts for later.
No account needed
Current date and time in your timezone, and exact arithmetic.
No account needed
Read and add income, spending, budgets and holdings, check live US stock prices and set price alerts.
No account needed
Look up encyclopedia articles. No account needed.
No account needed
Search issues and pull requests, list repositories, open issues.
Personal access token
Search your mail and prepare drafts. Never sends on its own.
Sign in with Google
Read your agenda and add events after you approve them.
Sign in with Google
Find files and read Google Docs as plain text.
Sign in with Google
Search files in your OneDrive and SharePoint.
Sign in with Microsoft
Search your Outlook inbox.
Sign in with Microsoft
PlannedSlack, Notion, Jira, Dropbox
Shortcuts follow your system: ⌘ and ⌥ on a Mac, Ctrl and Alt everywhere else. Press ? in the app to see every one.
Every record belongs to one account, and every query is scoped to it. Here's what lives where.
Finance prices and configured alerts run on the server. Each account connects its own Alpaca keys and private Telegram chat; Telegram handles messages sent there. A linked chat and enabled alert rule or close summary are required for automatic messages.
Create an account in a minute, or run Agent2Work on your own machine with Ollama.
