Aion 3.0 Mini with a signed-in browser: verify the API route

Aion Labs currently lists aion-labs/aion-3.0-mini in its hosted API model documentation. To use it for a signed-in browser task, first confirm the exact model is still in the live catalog and that the catalog marks tool calls as supported. Then verify your local agent client can use the Aion API and run local tools. Tabbit Browser with Tabbit CLI can provide the local browser connection. Stop at a named, human-reviewed draft; no Aion-to-Tabbit integration or model run was tested for this guide.

A Reddit browser-agent user recommends pausing before the final external action and reopening the result page. That is general workflow feedback, not Aion Mini evidence. Original discussion.

Key takeaways

• Aion Labs documents the hosted model ID aion-labs/aion-3.0-mini and a no-key model-list endpoint. Recheck that exact ID in GET /v1/models before a task.

• Aion 3.0 and 3.0 Mini are separate entries. Do not substitute the full model or a 3.5 variant.

• The API reference describes tool request fields, but this research did not confirm Mini’s live tool_calls capability. Treat it as unknown until the exact catalog row confirms it.

• Aion model access, local-client execution, Tabbit CLI reachability, and the website login are separate gates.

• Use a stable dashboard ID, keep a source URL for each observation, and stop before changing the dashboard.

Check the exact model before connecting the browser

The Aion API reference says API endpoints are under https://api.aionlabs.ai/v1/, compatible with OpenAI client libraries, and that GET /v1/models does not require a key. Other API requests require an Aion API key. Its current model page lists aion-labs/aion-3.0-mini with a 128K context and describes Mini as a multi-model system based on the DeepSeek family, aimed at narrative structure and mature or darker themes. Full Aion 3.0 is a different ID and is described as based on the GLM family. Aion 3.5 Mini is another model again. These descriptions do not establish dashboard extraction quality or browser-agent reliability. See the Aion Labs models page and API reference.

The API docs do not show a Mini sunset date, and I found no separate official retirement notice in the pages reviewed. This is a dated check, not a guarantee that the API will remain available. Query the official model list before use. A direct unauthenticated catalog request from this environment returned HTTP 403, so the API documentation was the confirmed hosted-availability evidence. Do not hard-code an unverified completion endpoint or assume the model supports tool calls.

Draft a dashboard change report

For a read-only task, compare a dashboard’s displayed value with its prior visible value or a second approved source page. Ask for a specific record ID, field name, current value, date shown, and page URL. A Reddit user described session state and checking results as unreliable parts of browser-agent work, and suggested pausing before the final external action. That supports a review checkpoint, not a claim about Aion. Original discussion.

Use this prompt only after the live catalog confirms tool_calls for the exact Mini ID and the client passes the local-tool check:

Use Aion Labs model ID aion-labs/aion-3.0-mini only if it is present in the live official model catalog and tool_calls is true for that exact ID. Use the installed tabbit skill and Tabbit CLI from this local agent. Read only [authorized dashboard URL] and [authorized comparison URL] in separate task tabs. For stable record ID DASH-318, report each requested field, displayed value, visible date, and source URL. Separate direct page text from your interpretation. Mark unavailable or conflicting values unknown. Do not edit, submit, assign, or refresh credentials. Prepare a browser-visible draft titled "Aion Mini dashboard delta — draft 01" with a list of items for human verification. Stop before saving or sharing. If any model, local tool, browser, or login gate fails, name the gate and return the source checklist instead.

A second Reddit thread says pages can show stale or partial state and recommends reading the displayed result after each step. Treat that as a useful check, not evidence that Aion Mini supports browser tools. Original discussion.

Use Tabbit Browser for the local step

Install Tabbit Browser and follow the Tabbit CLI quick start. The task tab is separate from your normal tabs, so you can continue working while the agent reads the dashboard. The project’s local fixture verified that a second named Tabbit CLI task could reuse a test login in the same browser environment. It did not test Aion access, model tool calls, or a third-party integration.

Some clients expose /tabbit; Pi uses /skill:tabbit; use the host’s documented skill discovery. Check the client’s Aion provider support and local command permission separately. If the model or client cannot invoke local tools, have it produce a comparison plan, then perform the browser step with an authorized local agent. See the Tabbit CLI guide, connect an agent to an existing browser, first-task checklist, Browser MCP versus CLI, local versus cloud browser agents, research with source links, and safe login reuse, and Codex as a local host example. That guide does not prove Aion compatibility.

Secholar.L

Growth & Content at Tabbit

Secholar.L writes hands-on guides about AI browsers and agent workflows, comparing tools with real research and citing verifiable sources.