Developers

Built to be inspected, not just trusted

Agent versions, run states, connections, and audit events are first-class, queryable objects. On-prem and cloud deployment topologies, credential isolation, and step-level observability are part of the product — not afterthoughts.

Architecture stance

Server-authoritative, multi-tenant, auditable by design

Every protected operation is authorized on the server. Every tenant-owned record carries an immutable workspace key. Credentials never leave the secret store in clear form. Runs are explicit state machines, not loosely coordinated flags.

Deployment topologies

Where Captiva can run

On-prem servers

Deploy inside your own network. Ideal when field controllers, industrial systems, or regulated data cannot leave the perimeter.

Private / hybrid cloud

Run the control plane where you already operate. Connect both internal systems and external SaaS without forcing a single network model.

Public cloud

Fully managed option for teams that want the same governance model without operating the infrastructure themselves.

Run state machine

Every run has an explicit, recoverable state

DRAFT → READY → QUEUED → RUNNING → WAITING_FOR_INPUT | WAITING_FOR_APPROVAL | SUCCEEDED | PARTIAL | FAILED | CANCELED. Step-level events are persisted so the UI can reconstruct history without parsing free-form logs.

Security model

What engineers actually care about

  • ▸ Tenant isolation enforced server-side on every query
  • ▸ Credentials stored separately; only references and masked metadata reach the application
  • ▸ Approval required for destructive, financial, production, or externally visible actions
  • ▸ Idempotency keys for retriable mutations
  • ▸ Full audit event history: who approved what, which version ran, what changed
Observability

Logs and evidence that survive the 2 a.m. call

When a run fails or waits for approval, the operator sees the exact step, the inputs that were used, the tool results, and the recovery options — retry, resume, fork, or cancel — without reconstructing intent from scattered log lines.

// fetch the current state of a run
GET /v1/workspaces/{id}/runs/4821

{
  "status": "awaiting_approval",
  "agent_version": "lead-router@1.4.0",
  "environment": "production",
  "steps_completed": 4,
  "steps_total": 6,
  "pending_approval": {
    "step": "crm_sync",
    "reason": "new_account_threshold"
  }
}

Ready to see how Captiva fits your stack?

No pricing tables, no sign-up forms. Talk to us directly.

Talk to us