Eidos

Stop your agent from undoing what you already decided.

Write down once what you’re building, and what it won’t do, in markdown your coding agent reads before it touches the code. No more starting from zero every session.

Free for macOS, Windows, and Linux. No account, no subscription.

A feature blueprint for a streaming app open in Eidos, with the AI Agent beside it asking three scoping questions.
The agent reads the spec, asks where it’s vague, and changes nothing until you answer.

Sound familiar?

The review passes, and the product is still wrong.

  1. It changes it back

    A choice you ruled out on purpose comes back next session, written with total confidence. Nobody notices until it costs something.

  2. Every session starts from zero

    You re-explain the project, paste the old plan, and grow CLAUDE.md, until nobody remembers where its rules came from.

  3. Specs rot

    A spec kit writes a long document for every change, often by the agent. Months later the specs and the code disagree, and a stale doc is worse than none.

How it fits

One blueprint per feature, read by every session.

With the software framework, every feature gets one file: why it exists, what it promises, and what it won’t do. Your agent reads it before it writes a line, in the app or through eidos instructions in any terminal.

Specs
One per feature, grouped by domain: Intent, Open Questions, Acceptance Criteria, Out of Scope. A micro variant keeps small ones small.
Frames
The product as a whole: architecture, audience, criteria, and market. Every spec is judged against them.
Tasks
The work, on a board. Each task points at its spec, so the spec never fills up with sprint state.
The specs collection as a table, grouped by domain (playback, identity, discovery), with each spec’s summary, status, and type.

What you will use most

The pieces that keep an agent honest.

  • AI Agent

    Your agent, with the spec as context

    Claude Code or Codex beside the page, with the open spec attached. /grill makes it ask before it writes, and your own slash commands share the prompts that work.

  • Backlog

    Work beside the spec

    Tasks point at their spec. The agent moves them, ticks their checklists, and notes what it learned.

  • Blueprints

    A check, not a vibe

    eidos check flags a spec missing a section or a property, on your laptop and on every merge request.

No app required

Works from any terminal, too

The desktop app is optional. The same blueprints work from the CLI, and one command points any agent that reads AGENTS.md at the workflow.

$ curl -fsSL https://eidosmd.com/install | sh
$ eidos install @BuildableWorks/software
$ eidos agents --write
✓ created AGENTS.md with the Eidos nudge
$ eidos check
✓ no findings

Start here

Install the software framework.

Frames for the product, specs for the features, and a micro variant for small ones. Pick it when you make a new project, or install it from a terminal.

An app, a service, or a system

Software@BuildableWorks/software

Frames for architecture, audience, criteria, and market, and specs grouped by domain.

  • Frames
  • Specs
$ eidos install @BuildableWorks/software

In the app: New Project, then pick it from the list.

See it in the Registry

Give your agent something to check against.

Download Eidos MD, install the software framework, and write your first blueprint before the next session.

Free for macOS, Windows, and Linux. Your files stay plain text in a folder you own.