Reduce operating scope
Each run stays understandable because roles and decisions focus on one bounded responsibility.
Define roles, steps, contracts, routes, approvals, and artifact expectations—then inspect execution, recovery, and evidence.
Inside the module
See the working surface, step by step.
Follow a definition into its live execution. Every process run has a manager AI agent observing progress, artifacts, recovery needs, and decisions within the process’s configured self-governance boundary.
Manager agent · Every run
One agent watches the whole run—not only the current step.
Every process run is paired with a manager AI agent. It observes the run across steps and subprocesses, tracks the evidence expected from each role, and keeps recovery and decisions connected to the process policy.
Self-governance defines the action boundary
Observe → recommend → coordinate → act within policy
The configured level, permissions, and approval gates determine how far the manager may go. It can always surface a problem; it may intervene only when the process explicitly allows it.
-
01
Observe the complete run
Read step state, role assignments, events, approvals, subprocess activity, and every required artifact from one operating context.
-
02
Recognize what is blocking progress
Distinguish a missing or invalid artifact from a failed step, a waiting approval, a role problem, or a decision that needs more context.
-
03
Coordinate recovery
Ask the responsible role for corrected evidence, route rework, and help restore the artifact chain instead of abandoning the whole run.
-
04
Help decide—or escalate
Support a decision and apply a permitted recovery only at the allowed self-governance level. Otherwise, explain the issue and ask the authorized person to decide.
Subprocesses
Split complex delivery without breaking its evidence chain.
A parent step can call a reusable subprocess for one bounded outcome. The child runs its own roles, steps, artifacts, events, and recovery, then returns its contracted result to the parent run.
01 · Parent run
Reaches a bounded unit of work.
The parent preserves the wider objective and waits at a clear process boundary.
02 · Subprocess
Executes with its own run state.
Its team, manager oversight, decisions, artifacts, and recovery remain inspectable.
03 · Contracted return
Hands evidence back to the parent.
The parent continues only with the output and artifact trail the subprocess promised.
Reuse proven delivery
A stable subprocess can serve several parent processes without duplicating its steps or controls.
Localize change and recovery
Tune, replace, retry, or recover the smaller unit while preserving its relationship to the parent evidence trail.
Run-specific staffing
Keep the process. Change the team that performs it.
At launch, select a different team or override individual role assignments while the reusable process definition stays the same. Every run records who actually performed each role.
Reusable definition
Same roles, steps, gates, and artifact contracts
Route sensitive work to private agents.
Select a team whose agents use secured local LLMs and approved local tools when project data must stay inside a controlled environment.
Run controlled comparisons.
Repeat the same process with a revised process version, team, agent, or model assignment, then compare artifacts, corrections, duration, tokens, and cost to measure improvement.
- 01Choose the reusable process
- 02Select a team or assign agents by role
- 03Review preflight readiness and estimates
- 04Launch a durable, separately measured run
What it covers
Focused functions. Shared project truth.
The module has a clear operating boundary while remaining connected to the people, agents, automation, and evidence around it.
-
Reusable definitions
Author global or project-scoped processes with criticality, autonomy, operating mode, roles, steps, branches, loops, and artifact expectations.
-
Visual topology and contracts
Build the process on a canvas while editing role and step responsibilities in focused panels.
-
Exact staffing before launch
Run non-mutating checks, resolve every required role to a specific person or AI agent, inspect readiness, and review preflight estimates before creating the durable run.
-
Governed agent roles
Resolve AI-owned roles through CRM / HR bindings and execute them through AgentFramework with permitted tools, required artifacts, and evidence expectations.
-
Progression gates
Require artifacts, valid outcomes, approvals, and legal state transitions before work advances.
-
Live run control
Monitor active and waiting parent/subprocess runs with current step, assigned agent, progress, event stream, details, and permitted control actions.
-
Cost and usage telemetry
Observe actual or estimated provider cost, token volume, elapsed time, event rate, and tool usage by history window, definition, run, and project context.
-
Durable recovery and audit
Keep waiting approval, blocked, failed, refused, rework, cancellation, escalation, recovery, files, tool receipts, and history as operating records.
How it connects
A process turns responsibility into execution state.
Projects provide scope, CRM / HR resolves roles, Agents perform governed work, and Workflows can handle bounded automation inside the run. Outputs and evidence return to the originating project.