SPEXR · public beta

The spec-based IDE where agents do the work and you keep command.

SPEXR puts the agent session at the centre and ties every change to a written spec. It runs the agent you already have installed, and its own AI features run on your machine.

Download Source on GitHub

Specs

Every change starts as a spec.

A spec is a Markdown file with a goal and acceptance criteria. It moves through seven steps, from Specify to Ship, and SPEXR shows where each one stands.

Hand a spec to the agent and its context goes with it. The plan comes back as tasks you can tick off, and a drift check flags code that no longer matches what the spec asks.

The Specs view: workspace progress and each spec with its workflow steps — Every spec in the workspace, and the step it is at.

A spec at the Implement step with its plan as a task checklist, two of five done — A spec mid-implementation, its plan ticked off task by task.

Darkfactory

Every agent session on one wall.

Darkfactory lists every local agent session, grouped by project. Each card shows whether the agent is working, idle, done or waiting for you, and a summary written on your machine.

Click a card to pick the session up. Find an old one by describing it, open the pull request or local page it produced, and chain tasks into a schedule where each one waits for the ones it needs.

The Darkfactory wall: session cards for the ledgerline project, one flagged as needing you, with summaries written on the device — Live state, and summaries written on your machine.

Darkfactory search for “the one about month-end dates” with three matching sessions, and a plant schedule of three tasks — Find a session by describing it. The schedule runs tasks in order.

A pinned session card with its summary, a terminal, and a browser showing the reconcile report the session served — A card opens its session, with a browser on what it produced.

Experts

The right expert for each step.

SPEXR ships seven built-in expert personas, such as Design, Review and Software Engineering. Add the ones a project needs, and the expert that fits each workflow step switches on when that step goes to the agent.

The Experts panel: Design, Review and Software Engineering added to the project, Software Engineering active, more in the marketplace — One expert is active at a time.

Memory

What the agent learns, it keeps.

Memory is plain Markdown notes, kept for you and for the project, and loaded into every Claude session. Move a note between the two as it proves useful beyond one project.

The Memory panel listing project, reference and user notes, each with a short description — Notes the agent loads in every session.

Smart search

Search by meaning, on your machine.

Ask in plain words and SPEXR ranks files by meaning and by keyword together. A local model writes a one-line description of each file. Nothing leaves your machine.

Smart search for “where do bank dates get parsed” with ranked files and one-line descriptions — Ranked by meaning, described on-device.

Also in the box

Git, and more than one agent.

Git is built in: stage, commit, push and branch without leaving the IDE. The local model writes the commit message from the staged diff.

Source Control with one staged file and a commit message the local model wrote from the diff — A commit message written from the diff.

How it runs

Your agent, your machine.

SPEXR calls no hosted model API. It starts the agent CLI you have installed and uses its sign-in.
Summaries, search and commit messages come from small models that run on your machine.
Open source under the MIT licence, for macOS, Windows and Linux.
Public beta: it works day to day, and it still changes often.

Download SPEXR Product summary