Agent Cloud is the control plane for agents that do real work inside a company: model routing, identity with RBAC/ABAC, policy checks before actions, credential brokering, spend tracking, human approval, immutable audit, and an instant kill-switch. Symbiosis products run their agents here so security and finance can approve production use.
Capability is rarely the blocker. Accountability is. Teams bolt a model API onto a secrets vault, track spend in a spreadsheet, and hope security signs off. Agent Cloud puts orchestration, governance, cost attribution, and (where policy allows) payments in one place.
Treat agents like staff with badges and corporate cards. They sit in a catalog, get least-privilege roles, versioned policy, live observation, budgets per cost center, and one-click termination. Those controls are how security and finance say yes.
When an agent can spend, the bar rises: merchant allowlists, wallets with hard caps, dual approval above thresholds, tokenized rails so agents never see PAN, receipt capture, GL coding on every transaction. A programmable ceiling beats a lecture about autonomy.
Registry · RBAC · policy · model router · wallets · HITL · kill-switch · hash-chained audit
20 capabilities, each one something you can point at during a security review.
Multi-model orchestration
Route each step across OpenAI, Anthropic, Google, xAI, Azure, Bedrock, self-hosted, and open weights by quality, latency, cost, residency, or policy - with failover, fallback chains, and per-task model pins.
Agent identity & RBAC/ABAC
Each agent is a first-class principal with roles, attributes, org/team hierarchy, and least-privilege scopes - no shared service accounts. SCIM/SSO sync with your IdP.
Registry & catalog
Every agent registered, versioned, owned, documented, and tagged by risk class; shadow agents have nowhere to hide.
Policy & guardrail engine
Declarative, versioned policies evaluated before every action and every payment: allow, deny, or route to human approval. Dry-run / simulate before deploy.
Credential & secret brokering
Short-lived, scoped credentials issued per action; agents never see raw API keys, DB passwords, or payment secrets.
Tool & connector hub
Governed connectors to ERP, CRM, email, Slack/Teams, ticketing, data warehouses, and custom APIs / MCP tools - each call policy-checked and audited.
Spend intelligence (model + tools)
Token/tool cost by agent, team, task, and model; budgets, rate limits, anomaly alerts, and rightsizing suggestions (cheaper model, cache, batch, or kill runaway loops).
Agent wallets & payment rails
Prepaid or funded wallets per agent/team; payment-gateway integrations (cards, ACH, payouts) so agents can pay approved vendors under policy - dual-control, merchant allowlists, hard caps, instant freeze.
Procurement & AP autonomy
Agents draft POs, match invoices, schedule payments, and settle within catalog/amount rules - with receipt ingestion and automatic GL coding.
Human-in-the-loop + multi-channel approve
Approval queues in-console plus Slack, Teams, email, or mobile push; dual approvers for money movement; structured deny reasons the agent can act on.
Kill-switch & blast-radius controls
Terminate one agent, a fleet, or all payment credentials instantly; sandboxes, environment locks (prod vs staging), and regional residency fences.
Immutable audit & evidence packs
Append-only, tamper-evident log of plans, model calls, tool calls, approvals, payments, and outcomes - exportable for SOC 2 / ISO / model-risk / finance audit.
Multi-agent trust & orchestration
Agent-to-agent calls carry identity and policy context (intersection, not union); supervisor agents, handoffs, and shared task graphs without privilege laundering.
Memory & context governance
Scoped memory stores with retention, redaction, and ACL boundaries so agents don't become a permissions bypass for sensitive data.
Eval harness & regression gates
Task suites, faithfulness checks, and autonomy-increase gates - no promotion to higher spend or write scopes without passing evals.
Observability & agent traces
End-to-end traces: plan → model → tools → policy verdict → payment → outcome; latency, error, and cost overlays for ops.
Schedulers, triggers & webhooks
Cron agents, event-driven agents (webhook/queue), and replay-safe idempotency for financial and write actions.
Template gallery & agent blueprints
Production-ready blueprints: invoice reconciler, vendor payer, support triager, procurement drafter, travel booker, cloud cost cutter - scopes and policies included.
Simulation & policy preview
Replay historical actions against a candidate policy; see what would have been blocked, approved, or paid before you flip the switch.
Compliance & residency packs
Control mappings, data-residency routing, retention policies, and evidence exports tuned for enterprise and regulated buyers.
Governed agents, in production
5 weeks
from build to security approval
For the first agent deployed under Agent Cloud governance.
The two agents built before it, without governance, never got through review at all. That is the difference the control plane makes.
94%
of agent actions auto-approved under policy
The remaining 6% routed to a named human for approval.
0
actions executed without a policy decision
Every action is evaluated before it happens, and every one leaves an immutable audit line.
61%
lower model spend after routing
Cheaper models handle the work that does not need a frontier model, with no measurable quality change across evaluation sets.
Spend caps have been hit four times. Every one was a runaway loop, killed automatically before it became an invoice.
MODEL ROUTER
Multi-provider orchestration with cost/latency/quality routing, fallbacks, and residency-aware selection.
GOVERNANCE PLANE
Identity, RBAC/ABAC, policy engine, credential brokering, approvals, kill-switch, audit.
SPEND OS
Model/tool FinOps: budgets, attribution, anomaly detection, and optimization suggestions.
AGENT PAY
Wallets, payment-gateway rails, merchant allowlists, dual-control payouts, receipt + GL automation.
CONNECTOR HUB
Governed tools/MCP/API connectors with per-action policy and audit.
FLEET OPS
Registry, traces, evals, schedules, templates, and multi-agent orchestration.
Investment is bespoke to estate size, compliance needs, and deployment shape - book a demo and we'll map the right fit.
STARTER
- → Registry & identity
- → Core policy engine
- → Model router (multi-provider)
- → Audit log
- → Kill-switch
- → Basic spend dashboards
BUSINESS
- → Everything in Starter
- → Approval workflows (multi-channel)
- → Credential brokering
- → Spend intelligence + suggestions
- → Connector hub
- → Agent wallets (capped)
- → Templates & schedules
ENTERPRISE
- → Everything in Business
- → ABAC & custom policies
- → Agent Pay (gateway integrations, dual-control)
- → Multi-agent trust
- → Eval gates & simulation
- → Compliance / residency packs
- → Private deployment options
Agent Cloud, answered
Does Agent Cloud work with agents we've already built?
Yes. Agents integrate via SDK or gateway: they authenticate as Agent Cloud principals and their tool/model calls pass through the orchestration and policy layers. Most integrations start with the gateway pattern - no rewrite of the agent itself.
Which model providers does it support?
It's a multi-model orchestrator by design - route across hosted frontier APIs and self-hosted/open weights. Governance applies to identity and actions; the router picks (or pins) the model per task under your cost, latency, quality, and residency policies.
Can agents really spend money?
Yes - under Agent Pay. Agents get wallets with hard caps, merchant/vendor allowlists, and dual-control above thresholds. Payments go through your payment gateway with tokenized credentials; agents never hold raw card data. Every payment is policy-checked, receipt-linked, and immutably audited. Freeze a wallet or kill an agent and payment ability stops instantly.
What happens when a policy blocks an action or payment?
The action is denied or routed to a human approval queue, the agent receives a structured explanation it can act on, and the event is logged immutably either way. Policies are versioned, so you can trace exactly which rule decided - including which wallet limit or merchant rule blocked a payout.
Is this a chatbot platform?
No. Agent Cloud is the operating platform for agents that do work - call APIs, move records, orchestrate models, and (when allowed) pay vendors. It's the layer that makes that safe enough for security and finance to say yes.
How does spend monitoring help beyond dashboards?
Spend OS attributes model, tool, and payment cost to agents and cost centers, then suggests concrete fixes: route this task to a cheaper model, enable caching, lower a rate limit, shrink a wallet, or kill a runaway loop - before finance sees the invoice surprise.
See it against your reality.
A 30-minute walkthrough with an engineer who builds it - your use case, not a script.