Custom AI agents
Agents that follow your own instructions, review segments and campaign results with the tools you allow and prepare drafts for your approval. They run on a schedule (cron) or manually…
Agents that follow your own instructions, review segments and campaign results with the tools you allow and prepare drafts for your approval. They run on a schedule (cron) or manually. Available on Growth and above while the AI assistant is enabled for the organization (ADR-0089).
Safety boundary
- The agent has no send, schedule, publish or approve tool. If the model asks for a tool outside the catalog (for
example
send_campaign) no call is made and the run log showsunknown_tool; a catalog tool that is not allowed showsnot_allowed. - Write tools only create drafts. A person applies the draft in the dashboard: a segment draft becomes a segment, a campaign draft becomes a campaign in draft status. The draft campaign goes through the usual pre-send check and approval flow.
- Read tools return aggregated numbers only (no contact ids, emails or phone numbers). Tool arguments and results are not stored.
- Draft text and the summary pass the banned phrase check.
Tools
| Tool | Type | Description |
|---|---|---|
list_segments | read | Segment names and estimated sizes (at most 25). |
campaign_performance | read | Sent, delivered, opened, clicked and converted counts of campaigns run in the last 7/30/90 days. |
delivery_summary | read | Delivery counts by channel and status in the last 1–30 days. |
create_segment_draft | draft | A segment proposal from a fixed rule list: clicked_7d, not_opened_7d, inactive_30d, event_7d. |
create_campaign_draft | draft | Push campaign copy (title ≤ 65, body ≤ 240); the audience is existing segments or a segment draft of the same run. |
At most 5 drafts are written per run.
Setup
Dashboard → Campaigns → Custom AI agents → Create an agent.
| Field | Description |
|---|---|
| Instructions | Task and limits (up to 4,000 characters). |
| Allowed tools | A selection from the catalog above. |
| Schedule | 5-field cron and time zone (app time zone when empty); at most hourly. Without it the agent runs manually only. |
| Limits | Steps per run (default 6, at most 10) and a monthly token cap (default 200,000). |
Scheduled runs only happen while the agent is enabled; Run now also works on a disabled agent.
Runs and log
Every 5 minutes the worker scheduler claims due agents and adds them to the custom-agent queue; because the AI
provider lives in the API process, the job is processed by the API consumer. No new run starts while the same agent has
a pending or running run. Run statuses: pending, running, succeeded, failed, skipped (AI disabled, quota or
credits used up, plan feature missing or monthly token cap reached).
API
curl -X POST "https://api.pushizo.com/v1/apps/$APP_ID/custom-agents" \
-H "Authorization: Bearer $PUSHIZO_API_KEY" -H "Content-Type: application/json" \
-d '{"name":"Weekly analyst","enabled":true,"instructions":"Review last week campaigns and prepare a win-back campaign draft.","tools":["list_segments","campaign_performance","create_segment_draft","create_campaign_draft"],"schedule":{"cron":"0 9 * * 1","timezone":"Europe/Istanbul"}}'Read endpoints (GET .../custom-agents, .../runs, .../drafts) need the analytics:read scope; writes and
POST .../{id}/run need admin. Applying and dismissing drafts happens in a dashboard session only. Node SDK:
hs.customAgents.
Conversational AI agent
An agent that answers customers automatically on WhatsApp, SMS, RCS and live chat, consults your systems (order status, appointments) and hands the conversation to a human when needed. Available on Growth and above…
AI site search
A search box embedded in your website: visitors search your own content and, when they ask a question, get a short answer with sources from the pages found. Available on Growth and above…