Advanced Mode

Webhooks & workflows

Two ways to take an agent beyond the builder: trigger it from the outside via a webhook, or chain several agents together in a workflow.

Webhooks — launch an agent from the outside

1. Generate a personal API token

Head to Settings, section Personal API token. Click Generate a token.

Shown only onceThe raw token (starts with fgt_) is displayed once. Copy it immediately and store it somewhere safe (a password manager, a secret manager). We only store its SHA-256 hash; there is no way to recover it afterward.

2. Find your agent's URL

On the page of a deployed agent (/deploy/[id]) in Advanced Mode, the Launch from the outside section shows a ready-to-adapt cURL recipe.

3. HTTP call

bash
curl -X POST https://www.forgent.me/api/agents/bp_xxx/run \
  -H "Authorization: Bearer fgt_a1b2c3d4..." \
  -H "Content-Type: application/json" \
  -d '{"message": "Generate the weekly KPI report"}'

Response (HTTP 202):

json
{
  "sessionId": "sess_01234",
  "agentId": "agent_56789",
  "environmentId": "env_abcdef",
  "message": "Session started. Query the Anthropic API to retrieve the results."
}

The webhook creates a session against the deployed agent, sends your message as the first event, and returns immediately. The run itself is handled by the platform — you get the results back through Forgent, with nothing to configure on Anthropic's side.

Regenerate / revoke

In Settings, you can:

  • Regenerate — invalidates the old token instantly. Any scripts using it will stop working.
  • Revoke — deletes the token without generating a new one.

Webhook V1 limitations

  • One token per account (regenerating overwrites the old one)
  • No HMAC signature on the incoming payload
  • No automatic retry / idempotency key
  • The caller must poll Anthropic for results — Forgent does not do pull-then-callback
  • No per-token rate limiting (inherits Anthropic's rate limit)

Workflows — chain several agents

The Workflows tab (under My Agents) lets you build a chain of N agents where the text output of agent i becomes the input of agent i+1. The browser orchestrates the sequence — each step waits for the previous one to reach session.status_idle.

Build a workflow

  1. Add agents one by one via the selector
  2. Reorder them with the ↑ / ↓ arrows
  3. Remove them with the ×
  4. Type your initial input
  5. Click Run the workflow

Execution

The page shows the steps with their status: Pending → Running → Done (or Error). The text produced by each agent is visible in a scrollable box. If a step errors out, the chain stops and the following steps stay pending.

Why on the browser sideVercel enforces 60 s per function on the Hobby tier. Some agents (web search, file analysis) take several minutes. The browser keeps a long-lived SSE connection open through the internal proxy, so we're not capped at 60 s. The downside: keep the tab open during execution.

Use cases

  • Research → Synthesis: one agent collects raw data from the web, a second one structures it into a report
  • Analysis → Writing: one agent analyzes a CSV, a second one writes an explanatory email to the client
  • Generation → Translation: one agent writes in French, a second one translates it into 3 languages

Workflow V1 limitations

  • No persistence — the workflow disappears on refresh
  • The user must keep the tab open
  • No retry on a failed step
  • Only the text of agent.message / agent.message_delta is passed downstream — files and tool outputs are not chained yet
  • No branching path yet (always linear)