stefanprodan/1ctx

1ctx brings people and AI agents together in shared projects. Give each
agent a model, instructions and tools. Work with them in team chats, run
them on schedules, or hand work to the coding agents you already run,
such as Claude Code, Codex or OpenCode.
Context stays with the project. Working memory keeps compact project
notes in context across chats and automations. The shared knowledge
base provides long-term memory that agents consult and update as needed.
Long conversations continue through summaries. Optional deciders use
models such as Jev to flag run results that need
human attention.
Agents run commands in a sandbox. You control their tools and network
access. The server, web UI and scheduler ship in one binary, with project
data kept on your server.
Highlights
- **Memory with provenance.** Every knowledge revision records who wrote
it and in which chat or run. See its changes, roll back, or undo a
project note. Automations pick up where the last run left off.
- **Automations.** Run an agent on a cron schedule in your time zone,
or start a run by hand. Set deadlines and let agents flag results that
need your attention.
- **Delegate to your coding agents.** Hand an external task to an agent
you run outside 1ctx. It connects over MCP with a personal access
key, claims the task, reads the linked knowledge, asks when it is
stuck and reports back on the task's timeline. The same endpoint lets
it search the project's knowledge and chats and update the knowledge
base.
- **Conversations that continue.** Long chats are summarized with recent
turns kept in context. Fork a chat or bring another agent into a turn
with @name.
- **MCP and skills.** Connect MCP
servers and install Agent Skills. Choose
which capabilities a chat or automation can use.
- **Sandboxed commands.** Agents use a sandboxed shell with built-in
commands and a virtual filesystem to work with project files. They can
call HTTP APIs with managed credentials.
- **Repository context.** Mount GitHub and GitLab repositories read-only
so agents can inspect code alongside the project's knowledge.
- **Web and visuals.** Agents can search the web, fetch pages and create
HTML or SVG visuals inside the chat. You control their network access.
- **Model providers.** Use OpenRouter, OpenAI-compatible APIs, Gemini,
OpenCode Go or Azure. Each agent has its own model and instructions.
- **Self-hosted.** Run the Bun-based binary, use Docker Compose, or