Capka
A self-hosted AI agent that produces files on your Linux server, with isolated chat workspaces and support for Ollama or cloud models.

Capka is a self-hosted AI agent for people and teams whose work ends in a document, spreadsheet or code patch. It works with uploaded files, runs code and returns editable outputs you can revise through conversation. The app and its Linux sandboxes run on your own server.
It can consolidate sales exports, check contracts against amendments and emails, convert documents, or research material online. Reports and spreadsheets remain part of the conversation, so follow-up requests change the existing files. Server-side tasks keep running after you close the browser, and chats retain their files between sessions.
Each chat has a separate workspace. Team members keep their own conversations and files, while admins manage roles, model access, usage and rules that allow, require approval for or deny agent actions. An audit log records what the agent does.
Scheduled jobs can produce recurring reports without a fresh request each time. Telegram provides access to the same files and conversation context as the web app. Reusable skills capture procedures and templates, while MCP connectors let the agent work with PostgreSQL, internal APIs and CRM services.
Capka is free, open source under AGPL-3.0, and runs on Linux with Docker. It needs at least 2 GB of RAM; 4 GB is recommended. It supports local models through Ollama and hosted models through Anthropic, OpenRouter or OpenAI-compatible APIs. Files stay stored on your server, but cloud models receive the content needed for a task. With a local model, that content can remain inside your infrastructure.