Give the next session a place to start.

1. Capture one clear request

Start with a non-sensitive example, without customer details or credentials. Create a radar for your project and add a blip with the result you want, the constraints, and what would count as finished. For example: “Draft three onboarding emails for my app. Keep each under 120 words. Save the drafts for my review; don’t send them.”

2. Give it a home

Keep the request and its discussion on that radar. Add a tag or change its triage state as the work moves forward. A radar groups the work; a blip holds the request and its thread.

3. Connect the tool that runs your AI

On My Radar stores the work and your agent’s definition. A tool such as Codex or Claude Code runs the AI. You need an active On My Radar account, MCP access enabled for your plan, and a compatible MCP client. Plans and availability shows the current proposals and whether they can be selected.

For Codex CLI, add the server and complete the sign-in flow:

codex mcp add onmyradar --url https://onmyradar.io/mcp
codex mcp login onmyradar

For Claude Code, add the HTTP server, then open /mcp to authenticate:

claude mcp add --transport http onmyradar https://onmyradar.io/mcp

Read the consent screen before granting access. The connection is limited by your account’s permissions and the scopes you grant. Recording outcomes requires write access.

4. Load an agent and the current work

Create a named agent in On My Radar’s Agent Repo, or use one you already have access to. Give it a prompt describing its role and limits. In Codex, replace “My Agent” below with its name:

Load My Agent from On My Radar using harness codex. Adopt its prompt, read the relevant briefs and memories, then read my Launch radar and the onboarding-email blip with its comments. Summarize the current request before working.

In Claude Code, use harness claude-code. Loading an agent brings its prompt, a memory index and available briefs into the session. Ask it to read the specific memories and briefs the task needs. Check the current blip and source material too; saved notes can become outdated.

5. Leave a useful outcome

When the work is ready, ask:

Comment on the original blip with the drafts, what you checked, any open questions, and the next step. Keep the blip open until I’ve reviewed the drafts. Save reusable lessons in the agent’s memory, separate from this project’s task record.

Review the result and confirm it was saved. A comment records the project outcome; agent memory holds only the reusable lessons you choose to save. In your next session, load the same agent and read that blip again. You’re returning to a recorded request and outcome, rather than relying on the previous chat being available.

If you get stuck

Use Support when signed in, or Help without signing in for access problems. Describe the step that failed and any error message. Do not include passwords, API tokens, payment details or private project content. On the signed-out form, save the private conversation link and recovery code so you can return to the request.

Before you start

Early access is opening by invitation. Join the waitlist if you don’t have access yet. Your AI tool has its own subscription, permissions and limits. On My Radar does not run it continuously or guarantee that it recalls every detail.

Connection references: official Codex MCP documentation, Claude Code MCP documentation, and On My Radar’s connection overview.