Skip to content

Business logic in every prompt.

Rationale links team decisions and their reasons to the code.

Your code never contradicts your business rules.

If an agent is about to, Rationale stops it before the edit.

Nobody has to repeat it.

What your team decided reaches every agent, from every person, on its own.

Teams that don't collide.

When two people are about to touch the same thing, they know before the merge.

Fewer tokens. Less rework.

Your agent gets it right the first time because it already knows why.

Every decision has an owner.

Who stands behind it, what was ruled out and when to revisit it.

Early access · Guided setup

Connect Rationale to your AI

One remote MCP server for all your tools. Pick yours:

For Rationale to capture decisions from your Claude Code and Codex sessions, also install the client on each laptop:

curl -fsSL https://rationalehq.com/install.sh | sh
Full installation guide

What we eliminate

Your code says what changed. Nobody kept why.

No more decisions undone in good faith.

A teammate's agent reverts what you decided three weeks ago. The tests pass. You find out on a Friday.

No more tasks that break business rules.

“EU customers' data never leaves the EU region” arrives as a constraint no technical choice overrides.

No more duplicated or conflicting work.

When two people, or two teams, touch the same thing, they know before they collide. Not at merge time.

No more tokens spent repeating context.

Your agent starts out knowing what was decided. Fewer explaining prompts, fewer failed attempts.

No more “why was it done this way?”.

The pull request keeps the diff. Rationale keeps the why, who decided it and what was ruled out.

How it works

Captured while you work. Given back when it matters.

See the full example
  1. 01 · Capture

    Nobody has to remember.

    Install once per laptop. After each session, the client extracts the decisions on that same machine.

  2. 02 · Record

    Decisions, not transcripts.

    Question, choice, reasons, what was ruled out and when to revisit it. Tied to its files and its task.

  3. 03 · Give context

    Type #task. Your agent already knows it all.

    What was decided, who is on it and what is left. In Claude, ChatGPT or Cursor, over MCP.

  4. 04 · Warn

    Before the edit, not in review.

    When an agent is about to touch a file a decision is anchored to, the decision arrives first.

Nobody has to remember.

$ curl -fsSL https://rationalehq.com/install.sh | sh
$ rationale init
✓ Signed in as sam@acme.dev · workspace acme
✓ Hooks added to Claude Code and Codex
✓ Capturing. Nothing else to do.
sam@payments-api · claude · #faster-webhooks

› make the webhook retries faster, 3 attempts is too slow

● Reading app/jobs/webhook_delivery_job.rb

● Edit app/jobs/webhook_delivery_job.rb · MAX_ATTEMPTS = 3 → 10

Rationale · a decision is anchored to app/jobs/webhook_delivery_job.rb

D-42 Retry policy for partner webhooks

Decided by Ana Ruiz, in her session, 3 weeks ago · confirmed

Choice
3 attempts with exponential backoff, then dead-letter.
Because
two partners rate-limit us at 5 requests per minute.
Ruled out
more attempts (gets us blocked), a fixed 1 s delay.
Revisit if
partners move to the v2 events API.

⏺ I stopped before changing it. Ana chose 3 attempts on purpose: two partners block us above 5 requests a minute (D-42). I can add jitter and a longer backoff within those 3 attempts, or you can ask Ana to revisit D-42. Which one?

Illustrative names and code. The warning has the real format agents receive.

What we believe

Decisions with reasons and a source.

See the choice, its alternatives and when to revisit it.

Illustrative team decisions

  • business

    EU customers' data never leaves the EU region.

  • product

    Trials end without asking for a card.

  • technical

    3 retry attempts for partner webhooks.

Traceable source

See its source, evidence and whether it was inferred or confirmed.

Linked to the work

Decisions stay linked to tasks, tickets, pull requests and files.

Visible coverage

Unconnected sources and unlinked projects are clearly marked.

Integrations

Works with your tools.

Claude CodeLive

Capture each session, with pre-edit warnings and #handle context.

CodexBeta

Capture and context in the CLI, ChatGPT app and IDE extension.

Claude, ChatGPT, CursorLive

Read and record decisions over MCP.

GitHubLive

Link decisions to pull requests, reviews and issues.

JiraLive

Turn tickets and epics into tasks with their decisions.

No trackerLive

Keep tasks in Rationale and ask agents to open them.

Linear, Slack, Gemini CLI…

Not built yet. We build in the order teams request integrations.

Security

Built for code you cannot leak.

Encryption, permissions and audit
  • Decision content, notes and handoffs encrypted at rest; TLS everywhere.
  • Jira and GitHub by OAuth only; each member sees only what their own account can.
  • Every read and confirmation audited, including each time our staff looks.
  • EU company, GDPR by default; signed client releases.
Read the full security page

What we receive and protect

  • Extraction uses your agent provider; Rationale receives the results.
  • No repository clones. Document reads require an authorized knowledge source.
  • Secrets are refused and redacted from quotes.
  • We do not train AI on your content.
«The pull request keeps the diff. Nobody keeps the why. I built Rationale for my own team first: several people, several agents, the same code every day.»
The founder · Funchal

Questions

What teams ask first.

Anything else: support@rationalehq.com

What does Rationale receive from our sessions?

Your agent provider processes session excerpts under your own agreement. Rationale receives extracted decisions, short evidence quotes and file references, not full transcripts. Decision content is encrypted at rest. Separately, an owner can authorize document reads through knowledge sources.

Which agents does it work with?

Claude Code today, with capture and edit-time warnings; Codex in beta. Claude, ChatGPT, Cursor and any client that speaks MCP can read and record decisions through our remote connector.

Do we need Jira or GitHub?

No. Connect them and tickets, epics and pull requests carry their decisions. Without a tracker, tasks live in Rationale.

How is this different from CLAUDE.md or a memory MCP?

Keep using those for instructions and memory. Rationale adds decision records linked to tasks and files, with reasons, evidence and revisit conditions. Capture hooks and MCP bring that context into supported agent workflows.

What does it cost?

We are in early access and onboarding teams one by one. Plans come soon; we will publish them before anyone is charged.

Where is our data?

On servers in the United States, run by an EU company under the GDPR. Our privacy policy lists every provider and what each one processes.

Give your agents the why.

Try a guided two-week pilot with three people and one repository.

Plans are published before anyone is charged.