RUNTIME GOVERNANCE · ENFORCEMENT · SECURE RUNTIME

Don’t let your agents go rogue.

Understand what your agents do. Control actions that cross the line. Run them securely, from experimentation to production.

See How Sentience Works Runtime Governance is free and open source.
THREE PRODUCTS · EACH STANDS ON ITS OWN

Understand. Control. Run securely.

Pick the one that fits where you are. The demonstration below follows your choice.

Free · Apache 2.0Explore →
Per governed workflowExplore →
Workflow + scoped infrastructureExplore →

Free and open source. Works on its own, no paid product required.

8,000+Downloads
457Last 30 days
Local-first
SENTIENCE GOVERNOR WORKS WITH
Claude CodeCodexLangChainLangGraphPydantic AI
ONE SCENARIO · ILLUSTRATIVE

One agent execution. Three Sentience capabilities.

See the same run with:

Sentience Governor evaluates each action against the declared scope and policy, flags deviations and keeps the execution record, with the objective as context. It is advisory: it does not block the action.

Explore Runtime Governance →
Runtime Governance · sess_7f3a2cbilling-agent2 deviations · advisory
DECLARED OBJECTIVE
Answer a customer’s billing question and update their support ticket.

Not part of the task: issuing refunds or emailing the customer.

DECLARED SCOPE · what it may access
Billing recordsbilling.records
Support ticketssupport.tickets
DECLARED OPERATIONS · what it may do
Read recordsread
Update support ticketsupport.ticket.update
WHAT THE AGENT DIDGOVERNANCE FINDING
Read billing history
14:02:09 · billing.get_history(customer=10577)
Within declared scope
Updated the support ticket
14:02:21 · support.update_ticket(id=SUP-4412)
Within declared scope
Issued a refund
14:02:33 · billing.issue_refund(amount=184.20)
Flagged · action proceeded
Refunds aren’t a declared operation.
OPERATION_UNDECLARED
Emailed the customer
14:02:45 · email.send(template=refund_notice)
Flagged · action proceeded
Emailing the customer is outside the declared scope.
SCOPE_DRIFT
2 of 4 actions went beyond what the agent declared. Both proceeded and were recorded.evidence saved locally · sess_7f3a2c.jsonl

Illustrative scenario. No live agent is running on this page.

DEPLOYMENT

Your infrastructure. Or ours.

Enforcement runs where your agents already run, or inside isolated infrastructure we provide.

YOUR EXISTING INFRASTRUCTURE
Where your agents already run

Works with your existing agent infrastructure. No migration required.

YOUR CLOUDAWSAzureGCPPrivate
Your agentControl pointallowedTools & data
SENTIENCE SECURE RUNTIME
Isolated infrastructure we run

Execution infrastructure we provide, with governance and enforcement integrated.

SENTIENCE RUNTIME
ISOLATED WORKSPACE
Your agentControl pointallowedScoped tools
OUTSIDE THE WORKSPACEYour other systemsPublic internetonly what you allow
One strong foundation for both.Either way, you pay per governed workflow. Secure Runtime adds isolated infrastructure, quoted to fit what you run.
See pricing →
OPEN SOURCE · APACHE 2.0

Start with open-source governance.

Sentience Governor is free and open source. Run it locally on an agent you already have and see what it intended to do versus what it actually did.

Governor is advisory: it records and flags. Blocking actions at runtime is Runtime Enforcement.
GitHub PyPI Docs Pydantic AI integration →
$ pipx install sentience-governor
01Install Governor
02Attach to your runtime
03Run the agent
04Inspect the record
No account · No API key · No telemetry by default
Make your agent govern itself.
Copy this prompt into Claude Code, Codex or your preferred coding agent.
Set up Sentience Governor to govern your execution.

Official package: https://pypi.org/project/sentience-governor/

* Read the package documentation and install the appropriate integration for this project.
* Configure governance around your declared objective, permitted scope, and actions.
* Verify governance is active and show me your execution record and any policy findings.
* Explain what is governed, what isn’t, and whether enforcement is advisory or blocking.

Don’t assume installation equals active governance. Verify it.
USE CASES · EXAMPLE SCENARIOS

Built for real agentic workflows.

THE SITUATION

A coding agent is asked to fix a failing test. Partway through, it edits the CI configuration and tries to push to main.

GovernanceRecords every file change and command, and flags the CI edit as outside the declared scope.
EnforcementA control point can hold the push to main until a person reviews it.
Secure RuntimeRun the agent in an isolated workspace while you give it more autonomy.
Read the use case: Coding agent governance →

Example scenarios for each capability applied.

TECHNICAL PUBLICATION

From Governor’s Log

Read Governor’s Log →
LATEST · Oct 9, 2026Sixteen Approvals for a CopyHow a coding agent and an AI architect turned a reversible copy into a gated migration, and why a human approval every four minutes is not oversight.Read on Governor’s Log →
Oct 1, 2026A Valve Is Not a ReadIntegrating Sentience Governor into Antibody, a self-healing building of 14 agents and a coordinator on Flower, and what physical actions exposed.Sep 28, 2026The Price of RememberingA hackathon build on Claude, Pydantic AI and Sentience Governor that kept an agent’s mission through context compaction, and measured the token price of remembering.

Bring accountability to your AI agents.

Runtime Governance

Free and open source. Install locally and see your first execution record in minutes.

Get Started Free