We’re hiring

Know what your
agents built

Blume turns agent work into outcomes with evidence, and one human decision. Agents plan, build and test. You review the evidence and approve the revision.

Codex Align archived ticket handling
Claude Code Separate catalog sync detection
Grok Bot Refresh billing settings flow

LESS CHECKING IN

Stop babysitting
terminals.

Blume follows every session for you.

FROM BRIEF TO MERGE

Track deliveries
end to end.

Blume follows every session for you. Each change, from the first instruction to the merged commit.

READY FOR REVIEW

Everything you need
to get work shipped

Blume gives you a record of everything the agent tested, ready for you to review.

Works with

+ any other harness

Three challenges

As agents produce more code, the hard part is keeping their work visible, trustworthy and reviewable.

Visibility

Know what every agent is doing and what is left before you ship. Today, that means piecing together terminals, chat logs and branches.

Three agent activity rows gathered in one window, with flowers on the left
Test evidence inspected with a magnifying glass, with flowers on the right

Trust

An agent saying “done” is only the beginning. You need evidence that the change works and meets your team’s definition of done.

Review load

Several agents can build at once. Your attention is still finite. Bring the work and its evidence together so you can make one clear decision.

Work cards gathered for one review decision, with flowers at the top left and right
The delivery model

The unit of work is the delivery

Agents run the loop. Blume keeps the record. You step in when there is a verified revision to approve.

One product change with its brief, task tree, agent runs, code revision, tests and screenshots, intact from the first instruction to production.

Interactive product preview · Illustrative data

What you get

Output, confidence, quality

The process tells agents what “done” means, and every claim arrives with its evidence.

Get started

Your first delivery starts here

Download Blume, connect your agents, and review the evidence.

  1. 01

    Connect your agents

    Install the plugin into Codex or Claude Code. Every conversation connects on its own.

    A connected agent terminal with a small flower and an olive cable
  2. 02

    Run real work as deliveries

    Your agents plan, build and test. See what’s left at a glance.

    A nested todo list with completed and remaining work
  3. 03

    Approve with evidence

    Review test steps and screenshots. Approve the revision you ship.

    Test evidence inspected with a magnifying glass before human approval
Vision · On the roadmap

Your product, understood before work begins

A shared, versioned domain model gives every agent the same language, rules and edge cases. Each session starts with your team’s context, not a blank slate.

Your product’s working knowledge One model, shaped by real work

  1. Language

    Entities, relationships and the words your team uses

  2. Rules

    Business invariants, permissions and boundaries

  3. Journeys

    Critical paths and their expected outcomes

  4. Risks

    Known edge cases across time zones and currencies

  5. Architecture

    Conventions and patterns agents must preserve

Reviewed decisions, findings and bug reproductions keep it current as your product changes.

Used throughout every delivery

  1. Plan

    Define the right outcome

    Write acceptance criteria in the language your team uses.

  2. Build

    Preserve what already works

    Follow existing business rules and architecture boundaries.

  3. Test

    Test the risks you know

    Catch permission gaps, currency mistakes and dates shifting across time zones.

The whole delivery

Agents write the code. Blume makes it yours.

Download Blume and know what your agents built.

We’re hiring Open roles