Hector is an A2A-native AI agent platform written in Go, aimed at teams that need complete control over their own AI infrastructure. Its core philosophy is "your agents, your infrastructure, your rules." It is designed as an open-source AI agent runtime, providing a single self-contained binary, one YAML configuration, and production-ready defaults, and can be deployed locally, in isolated network environments, or in any cloud. There are no external dependencies, no runtime interpreter, and no mandatory cloud account.
Key Features
Single binary, zero dependencies: a Go executable of about 30MB with a built-in visual Studio UI. Copy it to a server and run it, with no interpreter, virtualenv, or package manager required. It works out of the box on any Linux, macOS, or Windows host.
Autonomy and control by design: can be deployed locally, inside an isolated network, or behind a firewall. LLM keys remain on the user's machine, data does not pass through third-party infrastructure, and there is zero telemetry. MIT licensed and fully auditable.
Open standards, zero lock-in: fully implements agent interoperability based on the A2A protocol and tool connections based on MCP. LLM providers (Anthropic, OpenAI, Gemini, Ollama) can be switched with a single line of configuration. Investments are portable.
Configuration-driven operations: agents, orchestration patterns, RAG pipelines, guardrails, triggers, and notifications are all defined in YAML, can be version-controlled, and can be reviewed in CI. A fluent Go API can also be used for fully programmatic control.
SQL-backed persistence: sessions, tasks, and checkpoints are persisted to SQLite or PostgreSQL. It can survive restarts, supports failure recovery, and supports human-in-the-loop workflows through built-in state management.
Enterprise security and multi-tenancy: supports JWKS/OIDC authentication, prompt injection detection, PII redaction, tool sandboxing, rate limiting, agent visibility control, and full tenant isolation. All declarative, with no code required.
Quick Start
Specific commands for Windows (PowerShell) installation are not currently provided. After startup, open http://localhost:8080/ — the terminal will print an admin key; enter that key to unlock Studio and configure an LLM provider, create an agent, and start chatting. No configuration file is required.
Hector Studio
Hector Studio is the Web UI for designing, testing, and managing agents, embedded in the Hector binary. Just run hector serve and open http://localhost:8080/。
Studio features:
- Visual flow builder: drag-and-drop canvas, bidirectional YAML sync.
- Integrated chat: streaming responses, with a tool call tracing view.
- Resource management: configure LLMs, tools, and guardrails without editing YAML.
Studio is also embedded in release builds and served at the root path. For how to start the development build, see the Studio guide.
Core architecture
SQL-backed foundation: all runtime state is persisted to SQL (SQLite or PostgreSQL).
Multi-tenant architecture: a single Hector deployment can host multiple isolated applications through the Admin API (non-file-based configuration). Each application has: isolated agent configuration, independent session and state storage, independent vector store collections, and per-application JWT authentication. Applications, sessions, and queues can be managed through REST endpoints. --auth-secret is required for authentication.
Session management, queue management, agent capabilities, multi-agent patterns, instruction templating (dynamically resolving instructions from session state).
Automation and integration
- Scheduled triggers: run agents on cron schedules, with time zone support.
- Webhook triggers: invoke agents over HTTP, with payload transformation.
- Outbound notifications: A2A-compatible push notifications on agent events.
MCP protocol support
Native Model Context Protocol, supporting all transports.
Command tools and security
Enterprise-grade retrieval-augmented generation (RAG):
- Document sources
- Advanced retrieval
Deterministic and LLM-driven content protection:
- Input guardrails
- Output guardrails
JWKS/OIDC integration, simple token authentication, Prometheus metrics, OpenTelemetry tracing, configuration validation.
License
MIT (see LICENSE).
Hector is an A2A-native AI agent platform written in Go.
