Give Claude Code the project state its next session needs.

Claude Code can carry instructions and learned context across sessions. Ro is designed to complement that memory with authoritative Tasks, decisions, responsibility, waiting, review, and history.

Join early access Why the session is not the work

Claude Code already has memory. Durable project state is a different job.

Anthropic documents two complementary mechanisms: instruction files such as CLAUDE.md or AGENTS.md, and auto memory that Claude writes from learned preferences, corrections, and project context. Both are loaded as context for later sessions; Anthropic explicitly describes them as context rather than enforced configuration.

That memory can help Claude Code orient itself. It should not have to stand in for the authoritative record of what the business is doing: which Task is current, who owns it, what it is waiting for, which conclusion governs it, whether a result was accepted, and what another person or agent must do next.

Claude Code memory and durable project stateComplementary layers

Platform reliability / Project contextReady

CLAUDE CODE CONTEXT

Help the harness orient itself.

CLAUDE.md / AGENTS.md

Instructions, architecture, commands, conventions, and working guidance.

Auto memory

Preferences, corrections, useful patterns, and project context Claude retains.

bounded
assignmentprogress
and result

RO PROJECT STATE

Keep the work record authoritative.

Investigate intermittent failuresTask · Operations · Review

Why: Customer-facing reliability issue
Waiting: Production sample received
Conclusion: Stale connection reuse
Result: Patch and tests awaiting review
History: 2 attempts · attributable

Bring your own Claude Code environmentRo retains no model credentials

Use each kind of memory for the truth it can carry.

Ro does not replace Claude Code’s memory files or put another model between you and the harness.

Instructions

Keep coding conventions and project guidance in the harness context.

CLAUDE.md or AGENTS.md can tell Claude Code how the repository works, which commands to run, and which practices to follow.

Learnings

Let auto memory retain useful patterns and corrections.

Claude Code can accumulate local notes that make a later session more effective. Those notes remain valuable as agent context.

Project state

Keep accountable work in Ro.

The Task, related Notes, current Discussion conclusion, dependencies, responsibility, waiting, result, review, and provenance remain structured and inspectable outside any one session.

Handoff

Return the attempt to the durable record.

A configured worker launches the chosen harness against bounded work. Progress and results return to the Task so the next human or agent continues from shared truth.

Bring Claude Code. Ro remembers the work around it.

Ro contains no model inference and does not sell an agent brain. You own Claude Code, model access, credentials, tools, working directories, and execution environment. Ro supplies the durable operational contract around the attempt.

Ro is under active development. Claude Code is an example of a user-chosen harness, not a claim of a shipping formal integration or affiliation with Anthropic.

Current product facts checked against Anthropic: how Claude Code remembers your project and Anthropic: give Claude context with CLAUDE.md.

Claude Code: Instructions and recall
Ro: Authoritative project state
Execution: Owned by you
History: Survives the session