A project planner that works with Claude and other AI agents
Disclosure: we build LatticeRun.
What is a project planner that works with Claude?
A project planner that works with Claude is a task tool your AI agent can read and update directly, so you stop re-explaining your project at the start of every session. LatticeRun is a visual task tree built for this: an agent claims a branch before changing it, receives the project's memory, and reports what it built. You can start the planner on a free account, no credit card. Agents, Project Memory and the Project Plan are on Solo Pro, $12/month.
What is free, and what is on Solo Pro?
Be clear about the split, because it decides whether this fits you. Free account, no credit card. Free projects stay in this browser. Agents, Project Memory and the Project Plan are on Solo Pro, $12/month.
- Free account, no credit card. Free projects stay in this browser.
- On Solo Pro, $12/month. Agents, Project Memory and the Project Plan.
- The planner itself. Free includes five of the six views of your task tree: mind map, list, table, workstreams and Gantt (sprints are a paid view). Progress rolls up from subtasks to their parents, so you can see what is left to ship.
- Backup. Free projects are saved in your browser's local storage and are not sent to our servers. Export JSON or SVG as a backup. If you clear your browser data, free projects are lost, so export anything you care about.
How does Claude work with it?
Agents connect through a first-party MCP connector, which is part of Solo Pro, $12/month. It works with Claude Code, Kimi, Hermes or any MCP-compatible client, and claude.ai and Claude Desktop connect by sign-in (OAuth).
Once connected, the working loop is simple:
- Claim before you write. The agent claims a branch of the tree before changing it, and reports what it built when it is done. An agent holds one claim at a time.
- Get the context. When it claims a task, the agent receives the relevant project context: the standing brief, bounded Project Memory and a plan synopsis. We do not promise perfect recall; we promise the agent starts with the project's recorded decisions rather than a blank page.
- Report back. What the agent built is attached to the task, so you have a record of what happened.
How do you stop an agent going off-plan?
LatticeRun has three enforcement modes. In Strict mode, task writes outside the claimed branch are refused. In Guided mode, out-of-sequence work is corrected rather than refused. Agents can read the enforcement setting, but no agent tool can lower it.
What is Project Memory?
Project Memory keeps decisions, guardrails, architecture notes, checkpoints and handoffs with the project. Each record is labelled proposed, approved or verified, so a successor agent does not mistake a suggestion for a settled fact.
Can I review what the agent produces?
Yes. Artefacts give each task one review thread. Every section can be approved, rejected or sent back for revision with a note, and the agent sees your decision on its next claim. LatticeRun emails the agent's address once per artefact when you decide or comment; a missed email is late, never lost.
Is it a replacement for my memory tool?
Often, no. LatticeRun is a task tree and context layer, not an agent runtime: it does not provide or replace the agent you already use.
FAQ
Is it free?
You can start with a free account, no credit card. Free projects stay in this browser. Agents, Project Memory and the Project Plan are on Solo Pro, $12/month.
Do I need an account?
Yes. Create a free account to start.
Which agents work with it?
Claude Code, Kimi, Hermes, or any MCP-compatible client; claude.ai and Claude Desktop connect by sign-in.
Can an agent change settings to loosen its own limits?
No. No agent tool can lower the enforcement setting.
What happens to free projects if I clear my browser?
They are lost, so export JSON or SVG first.
Who makes it?
LatticeRun is a trading name of Sequel AI Ltd, made in the UK.
Start with a free account · Agent setup guide · MCP connector · Use cases