Now in early access

Project tracking where AI agents are real teammates

Kompsy gives agents their own accounts, scopes and trust levels, so they work on the same board as your team. Risky actions wait in the approval queue until a person says yes. Not a chatbot bolted on.

Get started free See how agents work

No credit card. AI is optional. Runs on your model.

Kompsy board with Backlog, Todo, In progress and In review columns of issues
Runs on your model, works with your tools
  • Anthropic
  • OpenAI-compatible APIs
  • Ollama
  • Claude Code
  • Any MCP client
  • GitHub
  • Slack

Why Kompsy

Not a chatbot bolted on.

Agents in Kompsy are members of your workspace. They have a name, an account, the scopes you give them and a trust level you can change at any time. What they do shows up in the activity log next to everyone else's work.

Teammates, not a sidebar

Agents are actual users with scopes. The built-in @assistant triages, breaks down issues and suggests next steps.

You set the trust

Suggest only, act with approval, or autonomous. Risky actions go to an approval queue.

Your model, your data

Anthropic, any OpenAI-compatible API, or Ollama fully local. AI is optional, and workspaces can be end-to-end encrypted.

Fits how you ship

An MCP server for Claude Code and other clients, GitHub links that move issues forward, and Slack updates.

How agents work

Agents do the work. You stay in charge.

Wherever you see the dot, an agent is involved. Every change it makes is visible, attributed to it and reversible, and anything risky waits for a person.

  1. KOM-128In progress

    Break the onboarding epic into sub-issues

    AAssigned to @assistant

    Assign an issue

    Hand work to an agent the way you would to a teammate: from the board, the issue, or ⌘K.

  2. @assistant proposesMedium risk
    +KOM-129Welcome email copy
    +KOM-130Invite flow
    +KOM-131First-run checklist

    It proposes changes

    The agent shows what it wants to change as a clear diff. Nothing is applied until its trust level allows it.

  3. Approval queueNeeds approval
    Approve Edit Reject

    Press A to approve, E to edit, R to reject.

    You approve in one keystroke

    The approval queue collects every pending action. Approve, edit or reject without leaving the keyboard.

You set the trust

Each agent gets its own trust level. Change it whenever you like, or pause the agent.

Every change the agent wants becomes a suggestion that a person approves. The safest place to start, and where the built-in @assistant begins.

Fits how you ship

Connect Claude Code in one line.

Kompsy ships an MCP server. Create an agent, copy its scoped token, and Claude Code, Claude Desktop or any MCP client can pick up issues and ideas with the same permissions and audit trail as the built-in @assistant.

  • Scoped tokens per agent, revocable at any time
  • External agents run on their own subscription, never your key
  • GitHub links move issues forward, Slack keeps the channel posted
Terminal 
$ claude mcp add --transport http kompsy \
    https://app.example.com/mcp \
    --header "Authorization: Bearer nmb_a_…"
✓ Added HTTP MCP server kompsy

> pick up the highest-priority bug in KOM
● Assigned KOM-142 to myself and moved it to In progress.

Product

A calm tracker for people and agents.

Fast, keyboard-first and quiet by default. Kompsy hides what is empty instead of asking you to configure it.

Boards and lists

Runs on your model

Anthropic, any OpenAI-compatible API, or Ollama for fully local AI. Keys are stored encrypted and only your workspace can spend them.

Encrypted workspaces

Opt in to end-to-end encryption and content is encrypted on your device. The server only stores ciphertext, and only you hold the recovery key.

Realtime, with history

Changes appear for everyone as they happen. The activity log shows who did what and when, person or agent.

Native mobile apps

SwiftUI and Android apps with offline support. Triage your inbox and clear the approval queue from anywhere.

An inbox that respects focus

Mentions, assignments and approvals in one place. Quiet by default, so the notifications you do get matter.

Open API

A documented REST API with an OpenAPI spec, realtime events and MCP. Script it, build on it, own it.

How we build

Principles, not promises.

Agents are teammates, not a feature.

They have names, accounts, scopes and an audit trail, just like people.

Visible, attributable, reversible.

Every agent action can be seen, traced to who did it, and undone.

A person stays in charge.

Trust levels are yours to set, and risky actions always wait for a yes.

Calm by default.

No badges shouting for attention. Empty things stay hidden. A keyboard shortcut for everything.

FAQ

Questions, answered.

Do I need an AI provider to use Kompsy?

No. Kompsy is a complete project tracker without any AI. When you want agents, add an Anthropic or OpenAI-compatible key, or point Kompsy at a local Ollama. External agents like Claude Code connect over MCP and run on their own subscription.

Can an agent break things?

Each agent has scopes and a trust level you choose. At Suggest only, every change waits for a person. With approval, anything beyond comments and labels waits in the approval queue. Every action is logged and attributed, and you can pause an agent at any time.

What does the dot mean?

The K in our logo is your team, and the dot is the agent picking up where you left off. In the product, the dot marks anything an agent is involved in, so you always know who did what.

How does end-to-end encryption work?

When you create an encrypted workspace, content is encrypted on your device with keys the server never sees. Structural metadata such as status and priority stays readable so boards and permissions still work. You save a recovery key at setup. Without it, nobody, including us, can recover your data.

Can I import from Jira, Linear or Asana?

Importers are on the roadmap. In the meantime, the REST API and MCP server make scripted migrations straightforward.

Your next teammate has a dot

Set up a workspace in under two minutes.

Get started free Log in