AgentGrid — git for AI agents

Registry, version control, and observability for autonomous agent fleets.

pip install agentgrid

Quickstart

from agentgrid import AgentGrid

ag = AgentGrid(api_key="ag_sk_...")  # or set AGENTGRID_API_KEY

ag.register(
    name="email-triage-agent",
    framework="crewai",
    model="claude-haiku-4-5-20251001",
)

That's it. Your agent is in the registry with a version, a diff history, and an audit trail.

CLI

# Register or update an agent
agentgrid register \
  --name email-triage-agent \
  --framework crewai \
  --model claude-haiku-4-5-20251001 \
  --version-tag v1.1 \
  --message "Tightened prompt — reduce hallucination rate"

# View version history
agentgrid versions --agent email-triage-agent

# Diff two versions
agentgrid diff --agent email-triage-agent --from v1.0 --to v1.1

# Rollback
agentgrid rollback --agent email-triage-agent --to v1.0 --reason "regression in v1.1"

# Deploy to staging
agentgrid deploy --agent email-triage-agent --env staging

# Latency + cost metrics
agentgrid metrics --agent email-triage-agent --last 24h

# Auth
agentgrid login
agentgrid whoami

Python SDK

from agentgrid import AgentGrid

ag = AgentGrid(api_key="ag_sk_...")

# Register
agent = ag.register(
    name="email-triage-agent",
    framework="crewai",
    model="claude-haiku-4-5-20251001",
    version_tag="v1.1",
    deploy_message="Tightened prompt",
    environment="production",
)

# Fetch
agent = ag.get_agent("email-triage-agent")

# Version history
versions = ag.get_versions("email-triage-agent")

# Diff
diff = ag.get_diff("email-triage-agent", from_tag="v1.0", to_tag="v1.1")

# Rollback
ag.rollback("email-triage-agent", to_version="v1.0", reason="regression in v1.1")

GitHub Action

# .github/workflows/deploy.yml
- uses: nath-abhishek/agentgrid-sdk/.github/actions/agentgrid@main
  with:
    api-key: ${{ secrets.AGENTGRID_API_KEY }}
    agents-dir: ./agents
    environment: production
    deploy-message: ${{ github.event.head_commit.message }}

Each subdirectory of agents-dir that contains an agentgrid.yaml file will be registered on every push.

What This SDK Does NOT Do

The SDK has zero business logic. It is a thin HTTP client and CLI that calls the AgentGrid cloud API.

• No policy evaluation — all governance runs server-side in agentgrid-cloud
• No audit storage — events are emitted async to the cloud; no local DB
• No agent runtime — the SDK never invokes your agent code
• No framework lock-in — works with any Python agent (CrewAI, LangGraph, AutoGen, custom)
• No cloud lock-in — self-host the API and point AGENTGRID_API_URL at it

If you're looking for something that runs inside your agent process and makes decisions, that's not this SDK.

Framework-agnostic

# CrewAI
ag.register(name="my-crew", framework="crewai", model="claude-sonnet-4-6")

# LangGraph
ag.register(name="my-graph", framework="langgraph", model="gpt-4o")

# Custom
ag.register(name="my-agent", framework="custom", model="llama-3.3")

Environment variables

Links

• Docs: docs.agentgrid.dev
• Cloud: agentgrid.dev
• Issues: github.com/nath-abhishek/agentgrid-sdk/issues