TeamSpec Open Source Tool

AgentHub

The place where agent developers define their agent teams. Write your TeamSpec-compatible config once, version it, and deploy it anywhere Forge runs — with a full history of every change and every execution.

One place to define, version, and deploy your agents.

Whether you're deploying one agent or an entire team of agents, AgentHub is where the specs live. Every agent is defined here, versioned here, and controlled here. Forge reads directly from AgentHub at runtime, so what executes is always exactly what the config says — no drift, no surprises.

Define once, deploy anywhere

An agent config in AgentHub defines everything: the model, system prompt, tools, permissions, and environment. Forge uses that config to run in dev, staging, or production — with the right constraints applied automatically. Write it once; it works everywhere.

Full history, zero guesswork

Every change to an agent config creates a new version. You can compare versions, roll back to any prior version, or pin a production environment to a specific version while dev moves forward. Drift and mystery are eliminated.

Self-hosted and private

AgentHub runs in your own infrastructure. Your agent configs — which contain your system prompts, your tool mappings, your permission structures — never leave your environment. Apache 2.0 licensed and fully auditable.

What AgentHub gives agent developers.

01

Config Versioning

Every save creates a new immutable version. Production spawns can be pinned to a specific version. A new config version never silently takes effect in production — promotion requires deliberate action.

Immutable versions Version pinning
02

Audit Log

Every change to a config — and every spawn event triggered from it — is recorded in the audit log with actor, timestamp, and diff. Full accountability for every action your agents take and every config change that shaped them.

Immutable Log Actor Attribution
03

Access Controls

Define who can view a config, who can edit it, and who can launch it. Role-based permissions at the agent level mean sensitive configs stay controlled even in large teams.

Role-based Per-agent Permissions
04

Environment Tags

Tag configs as dev, staging, or production. Forge respects environment tags when selecting configs, preventing accidental production spawns from dev configs. Promote deliberately, not accidentally.

Dev / Staging / Prod Environment Safety
05

Schema Validation

AgentHub validates every config against the TeamSpec schema on save. Malformed configs are rejected before they can be executed. You cannot accidentally deploy an agent with missing permission scopes or an invalid model reference.

TeamSpec schema Pre-deploy validation
06

API-First Integration

Every AgentHub operation is available via REST API. Forge uses it natively. Your CI/CD pipeline can push new config versions on merge. Your approval workflow can gate promotion to production automatically.

REST API CI/CD Ready

What lives inside an agent config.

An AgentHub config is the complete description of an agent or agent team. When Forge reads it, nothing needs to be inferred or guessed. Define it clearly once; run it confidently everywhere.

Identity & Metadata

  • Agent name and description
  • Owner team and environment tag
  • Version number and changelog note
  • Created / modified timestamps

Model Configuration

  • Model provider and model ID
  • Temperature, top-p, max tokens
  • Context window budget
  • Fallback model on failure

Prompt Configuration

  • System prompt (versioned with config)
  • Prompt variables with default values
  • Output format specification
  • Chain-of-thought instructions

Tools & Permissions

  • Allowed tool list with scopes
  • Read vs. write permission per tool
  • Human-approval thresholds
  • Data access boundaries

AgentHub + Forge: config and execution, connected.

AgentHub and Forge work together as the definition and execution layers of a complete agent operations stack. AgentHub owns what your agents are; Forge owns how and when they run.

AgentHub is the source of truth

Forge never stores its own copy of agent configs. It reads directly from AgentHub at spawn time, ensuring that what gets executed is always exactly what the config says — including the specific version selected.

Forge writes execution events back

When Forge spawns an agent, it records the spawn event in AgentHub's audit log — who triggered it, which version was used, and what the outcome was. AgentHub becomes the record of both config and execution history.

Other tools integrate the same way

Because AgentHub is API-first, any tool in your stack can read configs or write audit events. Your own internal tooling, CI pipelines, and approval workflows all connect through the same API that Forge uses.

Self-host AgentHub in your own infrastructure.

AgentHub is open source and runs entirely in your own environment. Your agent specs, system prompts, and tool configurations stay where they belong — with you.

Step 1

Install from GitHub

Clone the AgentHub repository. AgentHub runs as a lightweight service backed by a Postgres database. Docker Compose and Kubernetes manifests are included.

github.com/TeamSpecAI/agenthub →
Step 2

Create Your First Agent Config

Use the AgentHub UI or REST API to create your first agent config. Define the model, system prompt, tools, and permissions. AgentHub validates the schema on save.

Step 3

Connect Forge

Point your Forge instance at AgentHub's API endpoint. Forge will discover all configs you have permission to access and make them available for spawning immediately.

Learn about Forge →

FastBytes helps you get AgentHub running fast.

TeamSpec is open source. FastBytes.io offers implementation consulting — from initial setup through CI/CD integration and team onboarding. If you want your agent definitions and deployments running smoothly without building everything yourself, talk to FastBytes.