The shared-context layer for engineering teams and their agents.

Your coding agents make hundreds of decisions a day — and forget every one. Radar captures the reasoning, automatically, into one shared context your whole team and every agent can answer from. It lives inside your agents, not in a dashboard you'll forget to open.

Local-first. Nothing shared until you approve it.
every square is a decision your agents would otherwise forget
works inside the agents you already run
Claude Code Cursor Codex opencode pi
The problem

Your team has no collective memory.

You have Git for code. PRs for review. Linear for tasks. But the richest thing your team now produces — the reasoning flowing through your coding agents — disappears the moment the session ends. So the same work happens again. And again.

There's no Git of decisions. So your team flies blind on the most valuable thing it generates.

What it is

Five moves. Every decision earns its place.

Radar embeds into your coding agents and works in the background. It captures the decisions you and your team make — the claim, the reasoning, the why — and assembles them into one shared graph your whole team can answer from, human or agent.

// agent session, captured passively
$ radar watch --agent claude-code
→ decision detected: session storage → Postgres
→ reasoning extracted, queued for review
→ 0 forms filled, 0 tabs opened
Capture

Automatically.

No forms, no wiki, no ritual. Radar reads your agent sessions in the background and turns the reasoning into structured decisions. Capture costs you nothing.

// governance, just like Git
$ radar propose "sessions stay in postgres"
→ open to anyone on the team
$ radar merge --reviewer=alex
→ becomes team truth. read is free.
Share

With control.

Nothing becomes team truth until a reviewer merges it — proposing is open, merging is owned, reading is free. Just like Git. Share knowledge, never activity.

// months later, a new hire asks
$ radar ask "why is auth built this way?"
→ cited answer, in seconds
→ source: decision #214, 2026-03-11
Answer

Instantly.

Ask "why is auth built this way?" and get a cited answer in seconds, drawn from a decision made months ago. New hires ramp in an afternoon. Agents read from prior solutions instead of re-deriving them.

// agent about to contradict the graph
$ agent: switch sessions to Redis
⏸ halted — contradicts decision #214
$ radar resolve --reason "..."
→ one sentence becomes the record
Halt

Stay at the helm.

When an agent is about to contradict a decision your team already made, Radar stops it and asks you to engage. On the calls that matter, "proceed" isn't enough — you say why, and that becomes the record.

Govern

Wire it in once.

Connect, and Radar backfills the decisions buried in your past sessions — your graph is alive on day one. From there, it fills itself as your team works.

Stay at the helm. Stop the slop.

Most "human in the loop" is rubber-stamping fluent output nobody really read. Radar surfaces once, exactly when it matters, and stays silent the rest of the day.

Built for engineers

Not for everyone.

There are tools that promise to be a "company brain" — one memory layer for sales, finance, legal, and everyone else. Broad by design, bolted onto your CRM and your inbox. That's not this.

Company brain

Horizontal. Every department, one shallow memory layer, bolted onto your CRM and inbox.

Atomic Radar

One team, one workflow: technical teams building with coding agents. It speaks code, decisions, and repos — not deals and tickets.

Where we actually are

Early. Honest about it.

We're not going to invent stars or a leaderboard rank we don't have yet. Here's what's real today.

5
runtimes validated — Claude Code, Cursor, Codex, opencode, pi
Local
first, by default — nothing leaves your machine unasked
Design partners
only, for now — a small group, chosen deliberately
Your data, your infrastructure

Local-first. Built to stay yours.

Radar runs local-first, and the context your team builds is yours. Nothing leaves the machine unless you connect a workspace — per signal, on purpose. For teams who can't put their codebase reasoning in someone else's cloud, it runs entirely on your own infrastructure, and provably never leaves it.

We're picking our first design partners

Fewer decisions lost. Same work.

We're building this with a small group of engineering teams. If your team builds with coding agents and keeps losing the why — let's talk.

Become a design partner
or just join the waitlist →
FAQ

Questions, answered plainly.

Is this a project management tool?

No. Linear and Jira track what got done. Radar captures why — the reasoning behind decisions, from where the work happens.

How is this different from a "company brain"?

Those are horizontal — one memory layer for every department, bolted onto your CRM and inbox. Radar is built for one workflow: engineers building with coding agents. Deep on engineering instead of shallow on everything.

Is it surveillance of my engineers?

No. Radar captures decisions, gated by the author, and surfaces only what's published — never a live feed of who's doing what. It measures adherence to an agreed practice, not how anyone works.

Do my developers have to change how they work?

No. Radar is embedded in the coding agent and runs in the background. Capture is passive. The only time it surfaces is when an agent is about to contradict a standing decision.

Which coding agents does it support?

Claude Code, Cursor, Codex, opencode, and pi are validated today. The integration is host-agnostic — any agent that can run a command works.

Where does our data live?

Local-first by default, with the option to run entirely on your own infrastructure so it never leaves.