To map fields from a signed-in page with Ember-1, install Tabbit Browser and call Tabbit CLI from a local agent that can use Fireworks’ accounts/fireworks/models/ember-1 endpoint. Have the agent put the proposed values in a named draft you can check against the page before filling a test form. Fireworks model page
RevenueBench published the actual Ember-1 outputs alongside its source prompts and expected values. Its tasks pass text extracts to the model. Use that same comparison structure for the test form: record the source label, proposed value, and URL in the draft, then check each row in the browser. RevenueBench original runs
Key takeaways
Ember-1 is the model, not the browser connection.
Check model access, client skill discovery, and local CLI reachability separately.
Sign in yourself in the intended Tabbit Browser profile.
Keep the first task read-only and require evidence.
What the official model page confirms
The official model page documents the provider’s model identifier and listed capabilities. It does not certify that your account, agent client, or browser connection can use the model. For Ember-1 model background, see the provider’s model page and existing Tabbit setup guides. Do not confuse model support for tool calling with a working local browser route.
| Check | Proves | Does not prove |
|---|---|---|
| Model appears in picker | Client knows a model label | Account access or successful request |
| Short model request succeeds | Provider credentials and routing work | Tabbit CLI is connected |
| Tabbit skill runs | Client can reach local launcher | Site is signed in |
| Page values include IDs and URLs | Claims can be checked | Permission to make changes |
Prepare Tabbit Browser
Install Tabbit Browser and sign in to the target site yourself. Install the Tabbit skill where your local client discovers skills. Pi uses /skill:tabbit; some clients use /skills. Finding the skill and launching the CLI are separate checks. Use a separate task tab to keep your ordinary research tabs available. A cloud agent needs a separately supported local handoff; otherwise ask it to prepare steps for a local client. See the Tabbit CLI guide, browser connection guide, login reuse guide, and first-task checklist.
Map source fields to a test form
The task copies only named fields from a source page to a test form. Keep unclear values blank and require a source URL for every mapping. Keep the example separate from any customer or production data. The project’s local Tabbit fixture showed that a second task tab could read a signed-in test page after login. That verifies session reuse in the fixture only; it was not a model-specific run.
Use Tabbit CLI in the current Tabbit Browser profile. Read [SOURCE PAGE] and map only these fields to [TEST FORM]: [FIELD LIST]. Do not submit, save, or send. If a value is missing or unclear, leave it blank and explain. Create the draft “Ember field map” with source label, value, destination label, and source URL. Show it in the browser and wait for my review.Expected output is a concise evidence table: visible label, value, record ID where applicable, and source URL. Example values are placeholders, not a recorded model run. If a field cannot be checked on the page, leave it unresolved.
| Stage | Pass evidence | Stop and inspect when |
|---|---|---|
| Model | The intended provider returns a response | Model ID is missing or access is denied |
| Client | Tabbit CLI opens a task tab | Skill appears but launcher fails |
| Page | Expected account and content are visible | Login page or unexpected account appears |
| Result | Each field maps to a visible source | IDs, dates, or units disagree |
Keep a human checkpoint
| Action | Agent role | Human review |
|---|---|---|
| Read page | Extract named visible fields | Confirm the account and page |
| Explain | Draft a short summary | Check definitions and context |
| Fill test form | Enter reviewed values only | Compare each field with source |
| Submit or edit live data | Stop before action | Approve and perform the change |
Ember-1 is a candidate for a short extract-and-map workflow when the client can call Fireworks and reach Tabbit CLI. Keep the source and target separate, require URLs, and stop before submission. For the choice between a browser integration and a remote tool, read browser MCP versus CLI. The related Codex setup guide, browser safety guide, local browser-agent guide, and AI browser workflow guide provide additional context.
FAQ
Can the model open a signed-in browser by itself?
No. A local agent client and browser integration are required.
How do I invoke the Tabbit skill?
Pi uses `/skill:tabbit`. Other clients may show a `/skills` picker or their own documented entry point.
Will Tabbit CLI sign in for me?
No. Sign in yourself in the selected Tabbit Browser profile first.
Can I use a remote agent?
Only if it has a supported local handoff. Otherwise pass its checklist to a local client.
What should the first task do?
Read a small, authorized page and return source URLs and visible identifiers.
Should the agent submit changes?
Keep the first run read-only. Review any proposed changes before applying them.