Process squads

Process squads run unattended workflows over the API — no live caller. Each member needs the api capability. The runtime starts at the entry agent, runs tools, and can hand off along canvas edges to the next specialist (up to a step limit).

Create a process squad

Open Build → Squads → Create and choose Process.

Pick an entry agent with the api capability.

Add more api-capable members and draw handoff edges for the workflow.

Attach tools per agent (HTTP, integrations, send email/SMS, query, handoff edges). Avoid voice-only tools — they are not used in process mode.

Test from the squad Process Console, or call the Public API.

Invoke via API

Use invoke:write scope:

POST /api/v1/invoke/
Authorization: Bearer tq_live_…
Content-Type: application/json

{
  "target": { "type": "squad", "id": "SQUAD_UUID" },
  "input": { "order_id": "123" },
  "session_variables": { "region": "eg" }
}

You can also target a single agent with api capability (target.type: "agent"). Only process squads can be invoked as a squad.

Response includes run_id, status, output, and error. Full reference: Squads API — Invoke.

Session variables

Seed structured values at invoke time (session_variables) or extract them on handoff edges. Downstream steps see extracted keys in the process context. Manage expected variables from the squad Variables drawer (derived from edge extraction schemas).

Observe and webhooks

Process runs appear in Observe as process/API sessions (not phone calls). Cost views focus on LLM usage for process (no STT/TTS). Platform fee is a flat per-run charge (first minute included, then per-minute overage), the same single Platform total as voice/chat, not separate invoice lines. Subscribe to process-related webhook events to react when a run completes or fails. API-only agents have no Server URL: their results arrive through the organization webhooks (process.started, process.completed, process.failed). See API-only agents.

Pricing (platform)

Platform process fees (per-run include + overage) and Taliiq AI Gateway markup vary by subscription plan — see Billing → Compare plans in the dashboard (not duplicated here). BYOK provider usage is not billed as provider cost. Insufficient available balance (top-ups plus any included plan usage) rejects new invokes; in-flight runs still finish and debit.

Best practices

Prefer deterministic tools (HTTP, email) over interactive/browser tools
Keep steps short; use handoffs for clear phase changes
Pass IDs and structured fields in input / session_variables instead of dumping prose
Monitor run_id in Observe when debugging multi-step failures

By default each agent keeps its own LLM. To make every step use the same model, pick an agent under Settings → Brain (LLM). Only the LLM provider and model are shared; instructions and tools stay on each agent. See Squad settings.