Agent work
Agents carry a mission, capabilities, tools, memory, and a visible execution history.
operator documentation · current release
Know the control plane before you automate the work.
A compact field guide to agents, orchestration, identity, metered billing, and production boundaries. Start here; move into the console when the operating model is clear.
01
Platform map
One execution surface, six operating concerns.
Composed execution
Topologies coordinate several agents without hiding their individual decisions or outputs.
Governed boundary
Tenant scope, RBAC, approvals, quotas, and billing checks sit in the request path.
02
First governed run
Move from intent to reusable evidence.
- 01
Choose an agent
Start from a domain agent or build one with an explicit mission and capability set.
- 02
Bind context
Select workspace data, tools, provider credentials, and the execution topology.
- 03
Run with traces
Watch live events and outputs; every transition remains attributable and reviewable.
- 04
Approve or promote
Resolve human gates, inspect evidence, then reuse the run path that proved itself.
03
Operator paths
The shortest route to each job.
/command-deckCommand deckCurrent missions, approvals, and system posture./agentsAgent rosterInspect purpose, tools, memory, and execution state./orchestrationOrchestrationCompose sequential, parallel, hierarchical, or automatic topologies./observabilityObservabilityFollow traces, events, outputs, latency, and failure context./integrationsIntegrationsConnect OAuth providers, MCP servers, model credentials, and external systems./security-reviewsSecurity reviewsRead-only PR and repo audits: fan-out finders, adversarial verification, severity-ranked reports./settingsSettingsManage members, access, billing, account state, and platform controls.04
Security reviews
Hook a repository up to automated security checks.
The Security Reviewer runs a read-only fan-out audit over a pull request or repository: independent finders inspect auth, tenant isolation and IDOR, injection, secrets and config, and webhook and MCP surfaces; every finding is then challenged by an adversarial panel, and only findings that survive are reported with a verdict of reviewed or flagged. Every report is stored in history. Reviews never modify the code they inspect.
Run a review now
Open Security reviews and choose Run a review. Enter owner/repo and a pull request number, then queue it. The report appears in the list when the worker finishes. Reviews are dry-run by default, so nothing is posted back.
Connect GitHub for auto-review
Click Connect GitHub to install the App on your repositories. Pull requests on installed repos are then reviewed automatically, and the verdict is posted back as a comment and asecurity: reviewed or security: flagged label once you switch the reviewer out of dry-run.
curl -X POST https://api.agenticsuite.app/api/v1/security/review-reports/scan \
+ -H "Authorization: Bearer $ACCESS_TOKEN" \
+ -H "Content-Type: application/json" \
+ -d '{"repo":"owner/repo","pr_number":42}'Safe by default: read-only over the target, dry-run unless execute is requested, a global kill-switch, per-item holds, a scope denylist, and a per-organization daily review quota.
05
API & identity
Use the credential that matches the actor.
Browser operators
Password or configured social login issues a revocable access token and rotating refresh token. Organization roles and explicit permissions govern the console.
Service automation
Tenant service keys use as_test_ or as_live_ prefixes and carry narrow scopes. Personal tokens remain user-owned; webhook secrets sign delivery.
curl https://api.agenticsuite.app/api/v1/agents \
+ -H "Authorization: Bearer $ACCESS_TOKEN"Production API docs are closed by default. Enable them intentionally or use the checked-in API module guide for implementation contracts.
06
Billing model
Seats establish access. Usage captures the variable cost.
SubscriptionStarter · Growth · Scale
IncludedSeats + standard AI tokens
MeteredStandard overage + premium AI
Checkout creates an exact three-item Stripe bundle: seat quantity, standard usage, and premium usage. Included standard tokens are allocated before overage is posted. Premium usage is pay-as-you-go from the first token. Operators preview and commit multi-item plan changes atomically.
07
Production boundary
Green means evidence, not optimism.
- All backing services report ready and the database is at migration head.
- Provider probes answer with the configured funded launch models.
- Billing catalog, usage ledger, invoices, tax, checkout, and plan changes pass locally.
- Full browser smoke covers auth, private routes, advanced agents, topology, and billing controls.
- Postdeploy proof repeats public discovery, auth gates, private routes, and closed API docs.
Trust boundary
Technical controls are documented in the public trust ledger. Formal Terms, Privacy, DPA, certification, and SLA claims require the current operator-approved contract package.
Inspect trust ledger