One package.
Every agent governed.
managed-agents is the governance layer that sits between any AI agent framework and execution. Claude Managed Agents, LangChain, CrewAI, AutoGen, or your own harness — every tool call evaluated before it runs, every verdict sealed in your immutable audit trail.
Built for Claude Managed Agents.
Works with every framework.
Anthropic built the execution infrastructure. managed-agents governs what runs on it. Every Effect Claude yields passes through behavioral policy evaluation before the sandbox executes. One package. Any framework. Every tool call governed.
from nomotic_managed_agents import NomoticHarness
harness = NomoticHarness(
api_key="nm_live_...",
agent_id="nmc-...",
)
# In your tool execution loop:
await harness.govern(tool_name, tool_input)
result = await sandbox.execute(tool_name, tool_input)from nomotic_managed_agents import GovernanceDenied
try:
await harness.govern("write_file", {"path": "/etc/passwd"})
except GovernanceDenied as e:
print(e)
# Governance DENY — Scope violation (UCS: 0.21)Governance evaluates every Effect before the sandbox executes. Not a wrapper — an insertion point. DENY verdicts block execution entirely.
GovernedSession wraps Anthropic's session event log with governance records — every tool call linked to a sealed evaluation ID in your audit trail.
Every tool call is scored across scope compliance, authority, reversibility, behavioral consistency, and 10 additional dimensions. UCS score and full reasoning returned.
If the governance service is unreachable, tool calls proceed with a warning. Set fail_open=False to block on service errors for high-stakes agents.
When one Managed Agents brain delegates to another, Nomotic evaluates the handoff itself. The governance layer your orchestration has been missing.
How governance works
at the harness level.
Nomotic sits between Claude's tool call decision and your sandbox execution — the only point in the architecture where governance has the mechanical authority to stop an action before it happens.
Claude yields Effect
tool_name + tool_input sent to the harness loop
Nomotic evaluates
14-dimension behavioral policy check in <3ms via your API key
Verdict returned
ALLOW, DENY, or ESCALATE with full reasoning and UCS score
Audit record sealed
Cryptographic hash, dimensions, verdict — immutable in your trail
Your framework. Our governance.
Any combination.
Every integration uses the same Nomotic governance runtime. Same 14-dimension evaluation. Same audit trail. Same verdict format. The integration layer adapts to your framework — governance stays consistent.
LangChain / LangGraph
Drop-in governance node for LangGraph workflows. Wraps tool calls with behavioral evaluation. Works with all LangChain tools and chains.
pip install nomotic-langchain
Documentation →
CrewAI
Govern CrewAI tool execution across every agent in a crew. Intercepts role-based tool calls before execution with full verdict logging.
pip install nomotic-crewai
Documentation →
AutoGen
Governance middleware for Microsoft AutoGen conversation flows. Evaluates agent-to-agent messages and tool calls in multi-agent conversations.
pip install nomotic-autogen
Documentation →
HTTP Proxy
Language-agnostic governance via HTTP. Any framework in any language can evaluate actions through the REST API. Node, Go, Ruby, Java — all supported.
POST /api/v1/govern
API Reference →
MCP Proxy
Governance for MCP tool calls. The proxy intercepts every MCP request and evaluates it before forwarding to the server. Zero changes to your MCP setup.
pip install nomotic-mcp
Documentation →
OpenAI Agents
Governance for OpenAI's Agents SDK and Assistants API. Intercepts tool calls and function calls before execution. In development.
pip install nomotic-openai
Request early access →
Governance for no-code
and low-code platforms.
Beyond developer frameworks — Nomotic governance is coming to the platforms where non-technical teams are building and deploying agents today.
Zapier AI Agents
Connect Nomotic governance to Zapier's AI agent automations. Every action in a 6,000-app workflow governed before execution.
Make (Integromat)
Governance module for Make's visual automation platform. Evaluate agent actions at any point in a scenario before they execute.
n8n
Self-hosted workflow automation with Nomotic governance nodes. Evaluate AI actions inside n8n workflows with your own policy configuration.
Build your own integration.
The govern API is six fields.
Any language. Any framework. Any platform. If you can make an HTTP POST, you can govern your agents with Nomotic. The response is back in under 3 milliseconds.
Give your agents
an identity.
Birth certificates are free for all tiers. Start issuing in under five minutes.