How bhargavakandala/multi-agent-app-builder works
Multi-Agent App Builder turns a plain-English app idea into a planned, generated, tested, and reviewed project using local Ollama models. The CLI passes the request to the orchestrator, which owns stage order and artifact handling; specialist agents make model requests but do not call one another. Generated code is extracted from model responses and written into separate draft and final directories, alongside stage reports and run metadata. The workflow and integrations are documented; the graph uses sampled source for implementation links and does not infer additional runtime behavior from the example output.
Main entry points: orchestrator.py Build orchestrator · llm.py LLM adapter · agents.py Planner agent · __main__.py Command-line interface
Components 12
CLI and orchestration
- Command-line interface
dev_agent/__main__.py - Build orchestrator
dev_agent/orchestrator.py
Specialist agents
- Planner agent
dev_agent/agents.py - Developer agent
dev_agent/agents.py - Tester agent
dev_agent/agents.py - Reviewer agent
dev_agent/agents.py
Model access
- LLM adapter
dev_agent/llm.py
Build artifacts
- Code extractor
dev_agent/extract.py - Workspace writer
dev_agent/workspace.py - Build files
Other components
- App idea
- Local Ollama
Connections 20
- App idea → Command-line interface: provides idea
- Command-line interface → Build orchestrator: starts build
- Build orchestrator → Planner agent: requests plan
- Build orchestrator → Developer agent: sends plan
- Build orchestrator → Tester agent: sends code
- Build orchestrator → Reviewer agent: sends code and report
- Planner agent → Build orchestrator: returns plan
- Developer agent → Build orchestrator: returns code
- Tester agent → Build orchestrator: returns report
- Reviewer agent → Build orchestrator: returns review
- Planner agent → LLM adapter: requests completion
- Developer agent → LLM adapter: requests completion
- Tester agent → LLM adapter: requests completion
- Reviewer agent → LLM adapter: requests completion
- LLM adapter → Local Ollama: sends prompt
- Local Ollama → LLM adapter: returns response
- Build orchestrator → Code extractor: extracts code
- Code extractor → Build orchestrator: returns files
- Build orchestrator → Workspace writer: writes build outputs
- Workspace writer → Build files: creates files