1ctx

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 deploy on Kubernetes. Project data is stored in SQLite, and API keys are read from secret mounts.

Quick Start

You need Docker and an OpenRouter API key.

From a clone of this repository, write the secrets:

cd deploy/docker
mkdir -p secrets provision
openssl rand -hex 16 > secrets/user-admin.key
printf '%s' '<your OpenRouter key>' > secrets/provider-openrouter.key
chmod 644 secrets/*.key

user-admin.key holds the password of the admin user.

Add a provider and an agent in provision/instance.yaml:

apiVersion: config.1ctx.dev/v1
kind: Provider
metadata:
  name: openrouter
spec:
  wire: openrouter
  baseUrl: https://openrouter.ai/api/v1
  keyFrom: provider-openrouter
---
apiVersion: config.1ctx.dev/v1
kind: Agent
metadata:
  name: assistant
spec:
  provider: openrouter
  model: openrouter/free
  prompt: You are a helpful teammate. Answer directly.

Build the image and start it:

ONECTX_VERSION=dev docker compose -f compose.yaml -f compose.dev.yaml up -d --build

Open http://localhost:11236 and sign in as admin. Read the generated password with:

cat secrets/user-admin.key

The YAML in provision/ is applied at every start. After editing it, restart the container:

ONECTX_VERSION=dev docker compose restart

To run a release, omit compose.dev.yaml and replace <release-tag> with a tag from Releases:

ONECTX_VERSION='<release-tag>' docker compose up -d

License

1ctx is licensed under the Apache License 2.0. See third-party licenses for bundled dependencies.