codejunkie99/keel

Local-first macOS coding workspace with local Laya and optional Jev decision selection

What it solves

Keel is a local-first macOS coding workspace that integrates multiple coding agents (like Claude Code, Cursor, and DeepSeek) into a single native interface. It solves the problem of managing disparate AI coding tools by providing a unified workspace with task routing, session management, and a visual agent graph to track how agents and subagents interact with files.

How it works

Keel acts as a host that manages coding agents via the Agent Client Protocol (ACP). It uses a "decision" system to route tasks to the most appropriate backend—either through a local Core ML model (Laya), a hosted service (Jev), or a standard harness route (Normal). The host validates every routing decision before applying it, ensuring control over how tasks are executed.

Who it’s for

Developers using macOS who want a centralized hub for various AI coding agents and a visual way to manage complex agentic workflows and file changes.

Highlights

  • Unified Workspace: Combines a composer, transcript, terminal, and changes view in one native window.
  • Agent Graph: A visual representation of active sessions, subagents, and shared context files.
  • Intelligent Routing: Uses Laya (local) or Jev (hosted) to select the best route for fresh tasks.
  • Multi-Agent Support: Integrates providers like Claude Code, Codex, Grok, and an embedded DeepSeek loop.
  • Decision Tracking: Exports decision receipts as JSONL for later replay and analysis.
  • Crash Recovery: An append-only run journal to replay live streams and recover interrupted runs.

Related

  • Project
    Mathews-Tom/armory

    armory is a curated collection of 152 production‑grade packages (skills, agents, utilities, etc.) that extend Claude Code, Anthropic’s AI‑coding assistant. Each package is a self‑contained prompt or automation unit designed for real‑world developer workflows—planning, task decomposition, code review, security scanning, release preparation, and more. The repo provides a structured lifecycle (decision‑map → plan‑prompts → task‑decomposer → plan‑review → milestone‑runner) and a set of orchestrator agents (e.g., team‑lead, full‑stack‑builder) that can run these skills automatically. Intended for developers who rely on Claude Code in production, armory offers tested, opinionated workflows with explicit input/output contracts and failure‑mode handling.

  • Project
    Entelligence-AI/agentplay

    A local, privacy-focused tool that visualizes and replays the activity logs of AI coding agents, helping you audit what they read, edited, and where they encountered errors.

  • Project
    crawfordxx/xiaoma-durex-copywriter

    A Claude Code skill that applies a professional advertising methodology to generate witty, minimalist marketing copy and posters using a split AI-image and code-based layout pipeline.

  • Project
    huisezhiyin/sdd-riper

    SDD‑RIPER Light is a lightweight, repo‑native framework that adds a control plane around LLM‑based coding agents. It defines a clear loop—restating goals, creating a minimal spec, awaiting human approval, executing, validating with evidence, and syncing results back—so agents can modify code safely, auditable and recoverable. Four modular “skills” (light, strict, codemap, new‑chat‑ready) cover daily work, high‑risk tasks, unfamiliar codebases, and hand‑offs.