luiDots and GPT-6 Astra: OpenAI agents alongside n8n
Dots are always-on agents powered by GPT-6 Astra — not to be confused with GPT-6.1 Sol. See the right angle for n8n automations and hosting with lui.
The request often mixes names: “GPT Astra 6.1” and “Dots”. On OpenAI’s line, the correct framing is different. Dots are always-on agents announced on September 29, 2026, powered by GPT-6 Astra. GPT-6.1 Sol (API ID gpt-6.1-sol) is a cheaper model in the GPT-6 family — a different role from the announced engine behind Dots. In this article, lui separates the products and shows how that talks to n8n, without treating a desktop agent as a replacement for a versioned workflow.
What Dots are
Per OpenAI’s official announcement, Dots are agents with a cloud computer, their own browser, and connections to more than 4,000 apps. They learn preferences from feedback, can work in the background (with safeguards and permissions), and show up in ChatGPT, Slack, and Teams. The user sets approval rules; sensitive actions go through review. The initial rollout cites Pro, Business Premium, and Enterprise plans in eligible markets.
Operational summary:
- Product: Dots (agent)
- Underlying model (announced): GPT-6 Astra
- Promise: continuous work with personal/organizational context
- Control: app permissions, Custom Rules, auto-review
In form, this resembles other always-on agents on the market: great for open projects and follow-through; distinct from an automation conveyor belt with nodes, logs, and deploys.
And “GPT Astra 6.1”?
Do not treat “GPT Astra 6.1” as a single launch. In recent public communications:
| Name | Approximate role |
|---|---|
| Dots | Always-on agent in the ChatGPT ecosystem |
| GPT-6 Astra | Frontier model that powers Dots; in the API, ID gpt-6-astra (announced standard price US$ 10 / US$ 50 per million) |
| GPT-6.1 Sol | Lower-cost model in the API (gpt-6.1-sol); not the announced “engine” behind Dots |
If your team saw “Astra” and “6.1” in the same conversation, treat them as two axes: agent (Dots + Astra) versus an economical API model (Sol). In n8n, what you connect via an OpenAI credential is the API model ID, not the Dots app.
Dots versus n8n
| Dimension | Dots | n8n |
|---|---|---|
| Metaphor | Colleague with a desktop | Node orchestrator |
| Observability | Activity / in-product approvals | Executions, logs, versioning |
| Integration | Plugins and connected apps | 400+ nodes, webhooks, code |
| Data | OpenAI policies + permissions | Self-hosted: data on your VPS |
| Best for | Open-ended, multi-app work without APIs | Repeatable, auditable processes |
Neither “wins” alone. Dots (and similar agents) cover what is hard to API-ify. n8n covers what the company needs to prove happened.
Practical angles in n8n
1. OpenAI in the middle of the flow (API)
Use OpenAI models in n8n’s AI nodes for classification, extraction, and tool-using agents. Pick the ID by task cost, not by the Dots app brand. Simple tasks → economical model; long reasoning → more capable model — the same principle as our guide on AI automations.
2. Handoff via webhook
When a Dot (or a human in Slack) finishes an analysis, it fires an n8n webhook to write to the CRM, issue a note, or notify the team with deterministic validation. The agent decides; the workflow commits to the system of record.
3. n8n as the source of truth
Keep credentials, queues, and LGPD concerns on self-hosted. Keep the Dot off the path of data that cannot leave Brazil — or limit what the agent can see.
Checklist before “putting new AI” into the process
- Is the task open-ended (better for an agent) or closed (better for a workflow)?
- Do we need per-step logs and rollback?
- Which model ID does the API actually expose — and what is the price per million tokens?
- Where is the human in the loop?
- Can the n8n instance handle latency and memory from long-running agents?
lui’s role
A new model changes the AI node; it does not change the need for HTTPS, backups, and a stable instance. lui hosts self-hosted n8n in Brazil for the part of the process that must be yours: webhooks, credentials, and executions without the artificial caps of the vendor’s managed cloud.
If the team conversation mixed Dots, Astra, and “6.1”, align the vocabulary first. Then choose: an always-on agent for open-ended work, n8n for the conveyor belt — and the right OpenAI model only where interpretation pays for itself.
Start with an instance at lui.cx. See also Grok Bot and n8n and Claude Opus 5.5 on n8n.