← All tools

Tool · IDE

SPEXR

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

Public betausable, still stabilising

See SPEXR in action →

SPEXR inverts the usual layout: instead of bolting an agent into a sidebar, the agent session is the primary surface and your files and terminals orbit it. Specs are first-class artifacts: docs/specs/<id>.md becomes a plan, tasks, a diff and a PR, and a drift detector flags divergence. A two-scope memory (user + project) keeps what the agent learns.

It also ships local semantic search (hybrid keyword + embeddings, with one-line file descriptions generated on-device, no API calls, nothing leaves the machine), a full built-in Git provider, and Darkfactory: a wall of every local agent session across your config profiles, with live state, on-device summaries, and one-click resume.

Transparent about how it works: SPEXR does not call any hosted model API; it spawns your locally installed agent binary and reuses its stored credentials. Today that means a working agent CLI is a hard dependency; it runs Claude Code and opencode sessions side by side. It’s public beta: MIT, on macOS, Windows and Linux.

In use

  1. SPEXR welcome screen: the agent session on the left, and the seven-phase workflow from Initialize to Ship
    Welcome
  2. SPEXR specs view listing eight specs, each with its seven workflow steps and a Send to agent button
    Specs, each on its seven-step workflow
  3. SPEXR specs view beside the Experts panel, with Brainstorming, Design and Review personas in the project
    Expert agents
  4. SPEXR memory panel listing the project and feedback notes the agent loads every session, with baseline, user and project scopes
    Memory the agent loads
  5. SPEXR explorer search results ranked by relevance, each file summarised on-device, beside the welcome screen and the Experts panel
    On-device smart search