Adopting AgentPlat 1.1.0

AgentPlat 1.1.0 adds governed agent autonomy without automatically changing how existing agents work. All 66 packages are published and verified. Upgrade the AgentPlat packages your application already uses together to 1.1.0; installing every package is unnecessary.

Choose how each agent works

Mode	Typical use	Configuration
Instruction	A person or application requests a bounded task.	Keep an existing definition without `interaction`, or explicitly select instruction mode on a new revision.
Purpose	An agent evaluates work toward an owner-defined purpose within enforced limits.	Publish a purpose revision, connect governance and mission services, then qualify and activate it through owner commands.

Both modes can coexist. A purpose flag alone does not start a worker, grant tool permission or prove that a purpose has been achieved.

Keep the traditional workflow

Upgrade your existing coordinated dependencies and run your application tests.
Leave definitions without interaction unchanged. They retain instruction behavior and their historical definition digest.
Keep existing application authentication, approvals and effect controls.

For example, if your application already uses these packages:

npm install @agentplat/framework@1.1.0 @agentplat/sessions@1.1.0

Enable governed autonomy gradually

Start with one agent and a bounded purpose, such as assessing a support request, collecting evidence and proposing a response for approval.

For PostgreSQL, back up the database, stop older workers and apply migrations through 018 using the same explicit schema for all stores. Follow the migration guide.
Enroll the agent through AgentGovernanceServiceV1 with authenticated owner identity, explicit tool limits and budgets. The configuration starts suspended.
Create a definition revision with the appropriate interaction field:

// Fields on AgentDefinitionRegistry.createRevision input, not a complete agent.
interaction: { schemaVersion: 1, interactionMode: "instruction" }
 
// Use this alternative for a purpose revision.
interaction: {
  schemaVersion: 1,
  interactionMode: "purpose",
  governanceId: "support-governance",
}

Connect the execution controller, Room/Planner mission composition, assessors, persistent stores, scheduler and external-effect controls. Use requireGovernedExecution: true for a fully governed Room and qualify every participant. Publish the revision, then prepare and activate it through the authenticated owner commands.
Verify suspension, owner correction, restart and uncertain-effect reconciliation before connecting real business actions. Follow the detailed integration guide and runnable support demonstration.

To switch a governed agent back to instruction mode, select a compatible published instruction revision through the owner mode command, then qualify and reactivate. Keep limits, consumed budgets and origin history. Do not bypass governance by starting an older worker or removing enforcement records.

Optional Jev integration

Keep your existing assessor unless you choose Jev. The optional adapter setup (opens in a new tab) requires explicit server-side credentials and request/result mapping. It neither grants execution authority nor installs a scheduler. In the Rooms source example, the default PROPOSAL_REVIEWER=rules remains available; selecting jev is explicit.

The support demonstration uses deterministic assessors and simulated inputs. Validate your own model judgments and operational boundaries before production; release verification is not a claim of semantic reliability or AGI.

Support Adoption Pilot