ashlr-workbench
About
Local agent workbench bundling OpenHands, Goose, Aider, and ashlrcode against one local LLM, with ashlr-plugin MCP servers pre-wired.
Details
- Author
- ashlrai
- Categories
- Developer Tools, AI, Automation
Jump to
Setup
Install ashlr-workbench in your MCP client (Claude Desktop, Cursor, Windsurf, and others).
Repository: https://github.com/ashlrai/ashlr-workbench
Follow the installation instructions in the repository README, then restart your MCP client.
Your local coding-agent HQ — four agents, one local LLM, ten MCP servers, zero cloud dependencies.
A curated, fully-local toolbox that wires four open-source coding agents (OpenHands, Goose, Aider, ashlrcode) to a single LM Studio model (Qwen3-Coder-30B) and the same set of ten ashlr-plugin MCP servers. The workbench supplies one CLI (aw), one healthcheck, and one update path so the whole stack feels cohesive instead of four loose tools you have to babysit. No data leaves the machine unless you opt into a cloud fallback.
# 1. Clone the supporting plugin (provides the 10 MCP servers) git clone https://github.com/ashlrai/ashlr-plugin ~/Desktop/ashlr-plugin cd ~/Desktop/ashlr-plugin && bun install # 2. Clone the workbench git clone <this-repo-url> ~/Desktop/ashlr-workbench cd ~/Desktop/ashlr-workbench cp .env.example .env # then edit # 3. Start LM Studio, load qwen/qwen3-coder-30b, click "Start Server" # 4. Verify ./bin/aw doctor # 5. Use ./bin/aw start aider # interactive session in cwd ./bin/aw start openhands # autonomous Docker-based agent
ln -sf ~/Desktop/ashlr-workbench/bin/aw /usr/local/bin/aw
Pick the one that matches theshapeof the work — see "Usage examples" below.
┌─────────────────────────────────────────────┐ │ ashlr-workbench │ │ │ │ bin/aw → scripts/{start,health,update} │ └─────────────────────────────────────────────┘ │ ┌───────────────────┬───────┴────────┬──────────────────┐ ▼ ▼ ▼ ▼ ┌──────────┐ ┌────────┐ ┌────────┐ ┌──────────┐ │ OpenHands│ │ Goose │ │ Aider │ │ashlrcode │ │ (docker) │ │ (rust) │ │(python)│ │ (bun) │ └────┬─────┘ └───┬────┘ └───┬────┘ └────┬─────┘ │ │ │ │ └──────────────┬───┴────────────────┴───────────────────┘ │ same MCP surface ▼ ┌─────────────────────────┐ ┌────────────────┐ │ ashlr-plugin (10 MCPs) │ ←── │ LM Studio │ │ efficiency / sql / │ │ Qwen3-Coder │ │ bash / tree / http / │ │ -30B :1234 │ │ diff / logs / genome / │ └────────────────┘ │ orient / github │ └─────────────────────────┘ ┌────────────────┐ │ Ollama :11434 │ │ (fallback) │ └────────────────┘
All four agents talk to the same LLM and the same MCP tools, so behavior is consistent regardless of which one you launch.
- macOS 14+ (Apple Silicon recommended; Intel works but slower)
- Docker Desktop — for OpenHands
- LM Studio withqwen/qwen3-coder-30bloaded — primary LLM
- Bun≥ 1.1 — for ashlr-plugin MCP servers and ashlrcode
- Python 3.12+ — for Aider
- Node ≥ 20 / npm — fornpm install -g ashlrcode
- ~32 GB free RAM (Qwen3-Coder-30B in 4-bit needs ~24 GB live)
- ~30 GB free disk (Docker images + model)
- Optional:Ollamaas fallback LLM
- Optional:ghCLI for GitHub PAT (GITHUB_TOKEN="$(gh auth token)")
# Plugin (provides the MCP servers all 4 agents share) git clone https://github.com/ashlrai/ashlr-plugin ~/Desktop/ashlr-plugin cd ~/Desktop/ashlr-plugin && bun install # Workbench git clone <this-repo-url> ~/Desktop/ashlr-workbench cd ~/Desktop/ashlr-workbench cp .env.example .env # Per-agent installers (only run the ones you want) ./scripts/install-goose.sh # Goose via Homebrew pipx install aider-chat # Aider (or: pip install --user aider-chat) npm install -g ashlrcode # ashlrcode # OpenHands needs no install — aw start openhands pulls the image on first run
./bin/aw doctor # actionable diagnosis ./bin/aw health # full 13-point check
agents/openhands/config.toml # OpenHands runtime settings agents/openhands/mcp.json # MCP servers wired into OpenHands agents/goose/config.yaml # Goose source-of-truth (copied to runtime on launch) agents/aider/aider.conf.yml # Aider model + UX agents/ashlrcode/settings.json # ashlrcode overlay (XAI primary, LM Studio fallback)
Workbench-wide environment lives in.env(see.env.example).
For anything else, runaw doctor— it prints the exact fix for each problem it detects.
- Createagents/<name>/with the agent's config file(s).
- Createscripts/start-<name>.shthat launches it pointed at the workbench config and the LM Studio endpoint.
- (If it has a daemon) Createscripts/stop-<name>.sh.
- Add it to thecasestatements inbin/aw(require_agent,cmd_start,cmd_stop,cmd_status).
- Add validation lines for its config inscripts/healthcheck.sh.
- Update this README's agent-lineup table and architecture diagram.
SeeCLAUDE.mdfor project-wide conventions.
This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.
MCP bridge that lets Claude Code delegate heavy tasks to the Antigravity CLI (agy) — purpose-built tools, model routing with fallback, session continuity, and output truncation to save Claude's context and tokens.
Unified MCP server providing access to Claude Code, Codex, and Gemini CLIs through a single gateway. Features multi-LLM orchestration, persistent session management, async job execution with polling, approval gates, retry with circuit breakers, and token optimization. Install: npx -y llm-cli-gateway
Intelligent orchestration platform that routes tasks to the best AI model (Claude, Codex, Gemini, OpenCode) using LinUCB bandits, validates through consensus voting, and learns from outcomes. 29 MCP tools, dev pipeline, 8 memory backends.
Agent-native developer Q&A API with MCP + A2A endpoints for citations, job pickup, and answer submission.
Embeds intelligent guidance into AI workflows to organize development and ensure quality.
Open-source AI coding stack — bundled MCP servers, agent runtime, and developer tooling for shipping AI-native dev tools.
Async Parallel Antigravity for Codex & Claude Code
Run parallel, resumable, human-operable Antigravity CLI sessions from Codex, Claude Code, or any MCP-capable agent harness.
knowledge network for AI coding agents. Developers connect their agents to a shared pool of verified solutions — saving tokens, reducing debugging time, and getting better results. Solution authors earn when their work helps others.
Orchestrates multiple Claude Code agents across iTerm2 sessions, providing centralized management and inter-agent communication.
Multi-LLM Design and Build Team. Confer and create with a team of LLMs.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
