AgentHive is a free-first, Docker-first multi-agent orchestration platform for deploying, managing, observing, and executing specialized AI agents from a unified dashboard.
Designed for scalable multi-agent teamwork, AgentHive features dynamic pattern orchestration (Supervisor, Swarm, Router), speculative parallel LLM racing, a modular 40-tool catalog, and full observability integrations.
-
Multi-Agent Orchestration Engine (
MultiAgentEngine):- Supervisor Pattern: Decomposes complex user requests into discrete sub-tasks, delegates sub-tasks to specialized sub-agents, and synthesizes findings into a unified final answer.
- Swarm Pattern: Enables dynamic peer-to-peer control transfers using
[HANDOFF] agent_slug | {context}tags with execution loop safeguards (max_handoffs). - Router Pattern: Conducts fast intent classification to route queries directly to the optimal specialized agent.
- Auto Mode: Classifies intent and automatically selects the optimal orchestration pattern.
-
Speculative Parallel Multi-LLM Racing (
LLMRouter.generate_parallel()):- Queries multiple configured providers concurrently (Gemini 1.5 Flash, Groq Llama3, Ollama Local, OpenAI GPT-4o-mini, HuggingFace Llama 3.2).
- Utilizes
asyncio.wait(..., return_when=FIRST_COMPLETED)to return the fastest response instantly and cancel pending calls.
-
Categorized 40-Tool Catalog (
backend/app/tools/modules/):- 13 Infrastructure Base Tools:
file_tool,postgres_tool,redis_tool,minio_tool,loki_tool,code_tool,scraper_tool,translation_tool,youtube_transcript_tool,md_writer_tool,search_tool,pdf_tool,csv_reader_tool. - 9 Web Tools:
wikipedia_tool,arxiv_tool,rss_reader_tool,url_checker_tool,weather_tool,dns_lookup_tool,whois_tool,hacker_news_tool,github_repo_tool. - 5 Document Tools:
docx_tool,excel_writer_tool,pptx_tool,ocr_tool,json_yaml_tool. - 5 Text & NLP Tools:
sentiment_tool,text_summarizer_tool,diff_tool,keyword_extractor_tool,markdown_to_html_tool. - 5 Utility Tools:
calculator_tool,datetime_tool,image_metadata_tool,hash_crypto_tool,zip_archiver_tool. - 3 Developer Tools:
git_tool,sql_query_builder_tool,json_schema_validator.
- 13 Infrastructure Base Tools:
-
REST API Endpoint Extensions:
POST /api/orchestrate: Dynamic multi-agent pattern execution endpoint.POST /api/agents/{id}/run: Direct or delegated multi-agent run endpoint.
-
Full Observability & Traceability Stack:
- Grafana Loki (:3100) container log aggregation.
- Prometheus (:9090) endpoint and LLM latency scraping.
- OpenTelemetry distributed tracing spans.
- Grafana (:3001) visual dashboards.
| Component | Technology | Role |
|---|---|---|
| Frontend | Next.js + React + Tailwind CSS | Interactive dashboard & monitoring UI |
| Backend API | Python FastAPI | REST API & multi-agent orchestration |
| Multi-Agent Engine | Custom MultiAgentEngine |
Supervisor, Swarm, and Router pattern execution |
| LLM Router | Custom LLMRouter |
Speculative parallel racing & provider fallbacks |
| Database & Vector | PostgreSQL + pgvector | Persistent data, execution logs, & embeddings |
| Queue & Cache | Redis + Celery | Async agent tasks & job queues |
| Storage | MinIO | S3-compatible workspace document storage |
| Local LLM | Ollama | Zero-cost local inference fallback |
| Observability | Loki + Prometheus + OpenTelemetry + Grafana | Logs, metrics, tracing, & dashboards |
The XML architecture diagram is available in:
docs/architecture.drawiodocs/diagrams/multi-agent-architecture.drawio
Run the verification script to confirm valid XML schema structure and architectural keywords:
python docs/verify_drawio.pyRun unit and integration tests across multi-agent engine, speculative LLM router, and tool catalog:
pytest backend/tests/ -v- Start all containerized services:
docker compose up -d --build- Access core services:
- Frontend Dashboard:
http://localhost:3000 - FastAPI OpenAPI Documentation:
http://localhost:8000/docs - Grafana Monitoring:
http://localhost:3001(admin / admin) - MinIO Storage Console:
http://localhost:9001(agenthive / agenthive123)