Open Gravitre AI in Chat mode (/ai?mode=chat) for multi-turn conversational help across agents, workflows, connectors, and org data. Legacy /assistant redirects here.
Chat mode does not delegate tracked operator jobs — use Execute mode (/ai?mode=execute) for action plans and async jobs.
Start a conversation
- Sidebar → Gravitre AI (
/ai) → select Chat, or open/ai?mode=chat - Type a question in the composer at the bottom
- Press Send or
Enter(useShift+Enterfor a new line) - Watch the reply stream in — tool activity appears as chips above the message
- Use Stop to cancel an in-flight generation
Welcome quick actions suggest starter questions about Agents, Workflows, and Connectors.
Choose an intelligence mode
| Mode | Use when |
|---|---|
| Fast | Quick factual questions, low latency |
| Standard | Balanced everyday help |
| Reasoning | Multi-step analysis, tradeoffs, planning |
| Agent | Tool use against connected integrations when configured |
Modes change how deeply Gravitre thinks before responding — not every mode invokes connector actions. GIBE may rank retrieval and surface confidence when org learning data is available.
Choose a model
Open the model selector next to the mode control:
- Auto (recommended) — platform routing (may include GIBE rankers when trained)
- Featured models — GPT-5.5, Claude Sonnet, Gemini Pro
- Provider lists — additional OpenAI, Anthropic, and Google models
Model availability depends on your org configuration and provider keys.
When to use each Gravitre AI mode
| Situation | Use |
|---|---|
| Quick factual lookup in dialogue | Chat (/ai?mode=chat) |
| Multi-step assigned job with tracked plan | Execute (/ai?mode=execute) |
| Find a workflow by name | Find (/ai?mode=find) or Universal Search (/search) first |
| Execute connector writes | Execute or Workflows with Approvals |
Conversation sidebar
- New chat — fresh thread without losing history access
- Past conversations — resume prior threads; titles generate from first message
- Collapse the sidebar on smaller screens with the panel toggle
Messages persist locally up to a capped count; server-side conversations API stores durable history when enabled for your org. Chat turns feed GIBE learning when your plan includes Learning admin.
Org context
The org context pill shows what scope Gravitre AI uses (org name, environment hints). Click to review before asking sensitive questions.
Connector actions in chat
When integrations are connected, Gravitre AI may show:
- Tool chips — named tools invoked (read-only vs write)
- Connector action cards — proposed actions requiring confirmation
- Pending auth prompts — link to connect a missing Connector
- Readiness badges — Configured → Executable status before writes
Strict execution policy
Connector actions run only when integrations are linked and executable. The UI labels what was verified per turn — do not assume silent writes occurred.
Daily briefing
On load, Gravitre AI may show a daily briefing with bullets (errors, successes, warnings) linking to relevant pages. Dismiss or follow links to triage.
API
Streaming chat:
curl -s -N -X POST "https://gravitre.app/api/assistant/chat" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"messages": [{"role": "user", "content": "Summarize failed runs today"}]}'
Other endpoints:
GET /api/assistant/daily-briefingGET /api/assistant/org-contextGET /api/assistant/preferences
Browser clients use session auth; server integrations use API keys. See OpenAPI for request schemas and SSE response format.