ORCH — run multiple GPT4All-powered agents as a coordinated AI team with state machine + auto-retry

[oxgeneral]

Hey GPT4All community!

I built ORCH — an open-source CLI runtime for orchestrating multiple AI agents as a managed team. Using the shell adapter, it works with any local model, including GPT4All.

The use case: Run multiple GPT4All model instances as specialized agents — one for code generation, one for review, one for documentation. ORCH manages the task queue, enforces state transitions, and retries failures automatically.

# Register GPT4All-powered agents
orch agent add "coder" --adapter shell \
  --cmd "gpt4all run --model codestral-7b-v0.1.Q4_0.gguf"
orch agent add "reviewer" --adapter shell \
  --cmd "gpt4all run --model mistral-7b-instruct-v0.2.Q4_0.gguf"

# Build a local AI team pipeline
orch task add "Write authentication module" --assignee coder
orch task add "Review code quality" --assignee reviewer \
  --depends-on <code-task-id>
orch task add "Generate documentation" --assignee doc-writer \
  --depends-on <review-task-id>

# Real-time terminal dashboard
orch tui

State machine: todo → in_progress → review → done — reviewer must approve before closing. Auto-retry on failure with backoff. Zero infrastructure (YAML files only).

This is a fully local, privacy-first AI team setup — no cloud API keys needed if you run everything through GPT4All.

→ GitHub: https://github.com/oxgeneral/ORCH
→ Install: npm install -g @oxgeneral/orch

Has anyone here built multi-agent pipelines with local models? Would love to compare approaches!