██╗  ██╗███████╗ ██████╗████████╗ ██████╗ ██████╗ 
██║  ██║██╔════╝██╔════╝╚══██╔══╝██╔═══██╗██╔══██╗
███████║█████╗  ██║        ██║   ██║   ██║██████╔╝
██╔══██║██╔══╝  ██║        ██║   ██║   ██║██╔══██╗
██║  ██║███████╗╚██████╗   ██║   ╚██████╔╝██║  ██║
╚═╝  ╚═╝╚══════╝ ╚═════╝   ╚═╝    ╚═════╝ ╚═╝  ╚═╝

Production-Grade A2A-Native Agent Platform

Deploy observable, secure, and scalable AI agents in production—with zero code.

📚 Full Documentation → | Quick Start | API Reference

# Install
go install github.com/kadirpekel/hector/cmd/hector@latest

# Create configuration
cat > agents.yaml << EOF
agents:
  assistant:
    llm: gpt-4o
    tools: [search, write_file]
EOF

# Start server
export OPENAI_API_KEY="sk-..."
hector serve --config agents.yaml

Visit http://localhost:8080 for the web UI or use the CLI and REST API.

No YAML file needed—configure common use cases via command-line flags:

# Complete RAG system with Docling document parsing
hector serve \
  --docs-folder ./documents \
  --mcp-url http://docling:8000/mcp \
  --mcp-parser-tool convert_document_into_docling_document

This instantly enables document indexing, semantic search, and RAG capabilities. See Zero-Config Mode for more options.

• Observability - Prometheus metrics, OpenTelemetry tracing with Jaeger/Datadog/Honeycomb export, Grafana dashboards
• Security - JWT authentication with JWKS (Auth0/Keycloak/Okta), agent-level security schemes (Bearer, API key), command sandboxing
• Distributed Configuration - Hot reload from Consul/Etcd/ZooKeeper, zero-downtime configuration updates
• Rate Limiting - Multi-layer time windows (minute/hour/day/week/month), token & request count tracking, per-session or per-user scoping, SQL or memory backend
• Session Persistence - SQL-based storage (SQLite/Postgres), cross-session memory continuity, conversation history retrieval
• Human-in-the-Loop - Tool approval workflows, async HITL with state persistence (survives restarts), A2A Protocol compliant (TASK_STATE_INPUT_REQUIRED)
• Checkpoint Recovery - Crash recovery, rate limit resilience, long-running task support, event-driven and interval-based strategies
• TLS/HTTPS - Built-in TLS support for A2A server and vector stores
• Health Checks - Kubernetes-ready liveness/readiness probes

• Memory Management - Working memory strategies (buffer window, summary buffer), long-term memory with RAG, vector stores (Qdrant, Pinecone, Weaviate, Milvus, Chroma)
• Reasoning Engines - Chain-of-thought (iterative reasoning with tool execution), Supervisor (multi-agent orchestration and task decomposition)
• Tools - 10+ built-in tools (execute_command, write_file, read_file, search_replace, apply_patch, grep_search, search, evaluate_rag, todo_write, agent_call, web_request), MCP protocol support (150+ integrations via Composio), gRPC plugins for custom tools
• Multi-Agent Orchestration - Supervisor reasoning engine, agent_call tool, A2A-native federation, external A2A agent integration
• Streaming - Server-sent events (SSE) for real-time responses
• RAG & Semantic Search - Document stores with automatic indexing, advanced search modes (hybrid, multi-query, HyDE), LLM-based re-ranking, multiple embedder support (Ollama, OpenAI, Cohere)
• LLM Providers - OpenAI (GPT-4o, GPT-4o-mini), Anthropic (Claude Sonnet 4, Opus 4), Google Gemini (Gemini 2.0 Flash), Ollama (qwen3), custom providers via gRPC plugins
• A2A Protocol - 100% v0.3.0 compliant, agent discovery, standardized messaging and streaming, federation support

Complete documentation, guides, and examples available at gohector.dev:

• Getting Started
• Configuration Reference
• Production Deployment
• API Reference
• Core Concepts

AGPL-3.0 License. See LICENSE.md for details.