Skip to main content
Your agent can write the code. What it can’t do is know why your team ruled out that exact approach six months ago. That kind of context is real, but it lives scattered across chats, pull requests, and people’s heads, and it goes stale the moment the code moves underneath it. It is rarely just the code. The customer you spoke to last week, the decision that ended a long debate, the pipeline you are halfway through, the reason a project matters at all: each of those lives in a different tool, and you become the connective tissue between them. You re-explain the same state to the next teammate, the next agent, and the next session. The work gets done, and the context that made it possible quietly disappears. Driftless is one workspace for a team and its agents. Instead of letting that state scatter, it keeps three kinds of it together: the Knowledge you have reviewed and want to rely on later, the Operations you run day to day like customers, signals, and pipelines, and the Work in flight with its goals, criteria, and definition of done. Agents read and act on that shared state through permissions you set, and people stay responsible for what the company treats as true. No single one of those is the point. The value is the loop between them: a signal becomes context, context becomes work, work produces an action, and the action leaves behind reviewed Knowledge for whoever comes next. A message from a customer can become an operational update, open a piece of accountable work, hand an agent the context it needs, and leave a learning the team has actually approved. The Driftless loop walks through the whole model, and coding agents are the most concrete place to start.

​
Connect in about a minute

Same workspace whether a human or an agent reaches it. Install the CLI, point your agent at it, and you’re in.
1

Install the CLI and log in

npm install -g @driftless-sh/cli
driftless login
Running this in CI? Pass a key directly instead: driftless login --key drift_your_api_key_here. You can create one at driftless.icu under Settings, then API Keys.
2

Teach your coding agent

driftless install-skill
This drops AGENTS.md and CLAUDE.md into your repo. From then on, any coding agent that opens the repo already knows the rhythm: read what the team knows before editing, write back what it learned after.
3

Connect an AI client over MCP

Point ChatGPT, Claude, or Cursor at the Driftless MCP endpoint:
https://api.driftless.icu/mcp
Claude and ChatGPT handle the OAuth handshake themselves, so you just paste the URL. For the CLI, agents, and CI, send your key in the X-API-Key header instead. The full setup lives in MCP & OAuth.
Wiring an agent by hand? It really comes down to two moves: read before you edit, write back after you learn. Drop the whole loop in one prompt into any agent, or follow Agent setup for your specific client.

​
Explore the docs

How it fits together

Knowledge, Operations, and Work, and the loop that turns a signal into reviewed Knowledge. Start here for the whole model.

Connect

Every way a human or an agent plugs in: the CLI, agent skills, MCP, the GitHub App, Notion, and self-hosted agents.

Concepts

The Knowledge plane up close: workspaces, areas, topics, governance, and how drift keeps context honest.

Work & collaborate

Projects, collections, notes, and comments: the Operations and Work sides, where the doing actually happens.

Agents & automation

The Auditor’s PR comments, the server-side Audit and Architect agents, and the auto-pull hook.

CLI reference

Create, read, update, and anchor context topics from the command line.

API reference

The REST surface. Base URL, auth, the error envelope, workspaces, repos, topics, and integrations.