Muse Wonder and Muse Wonder Full share a name, but they are not the same access route. Muse Wonder has the API ID muse-wonder-1.0 and can be used with a local agent that Meta documents for Codex or Claude Code. Muse Wonder Full is available only inside Muse Code on eligible subscriptions; Meta says setting API reasoning to max does not select Full. For signed-in browser work, use the API model only after confirming your local client can run Tabbit CLI, or hand a Muse Code draft to a separate local agent for execution. A Reddit user described site blocks with Muse's own browser, which is a reason to verify access before planning around it. Meta availability docs · original community discussion
Key takeaways
M023 Muse Wonder Full and M024 Muse Wonder should be covered by one browser-use guide because the reader's task is the same: use a Muse model with an authenticated page. Their access routes still need separate treatment.
muse-wonder-1.0is the API model ID. Muse Wonder Full has no Model API ID and is Muse Code-only.Meta documents connecting Muse Wonder to Codex or Claude Code. In Codex, use the
/skillspicker to discover the installed Tabbit skill; confirm local shell access separately.Muse Code's model and subscription access do not prove it can invoke the local Tabbit CLI. For Full, hand the reviewed task description to a local agent unless your own setup verifies a local bridge.
Keep the browser job read-only until you inspect a named draft against the visible source.
Identify the exact Muse route
| Name shown | Verified access | Browser workflow decision |
|---|---|---|
| Muse Wonder | Meta Model API (muse-wonder-1.0) and Muse Code. Meta documents use with Codex or Claude Code. | A local Codex/Claude Code installation can be considered if it can select the API route and run local tools. Test both. |
| Muse Wonder Full | Muse Code only, on the Power Usage and Maximum Usage subscription plans. | Do not assume an API endpoint or local CLI bridge. Have Muse Code draft the steps, then hand them to a local Tabbit-enabled agent. |
API reasoning max | Highest reasoning setting for muse-wonder-1.0. | It does not switch the request to Muse Wonder Full. |
Meta's documentation currently says both variants are unavailable in South Korea. Muse Code also has Standard and Contributor modes; Meta says Contributor mode allows content to be used for training and can be turned off in settings. Check current access and data settings before sharing page text. Muse Wonder API documentation · Meta model list
Set up the local browser path
For Muse Wonder API, use a local client Meta documents, such as Codex, and verify the model appears in its current provider/model selector. Install Tabbit Browser and follow the Tabbit CLI guide. In Codex, open /skills and confirm the tabbit skill is available. Then use the skill through Codex's supported skill flow; /skills is the discovery interface, not a universal model command. Check that Codex can launch the installed Tabbit CLI from the same computer. Codex setup details · connecting an agent to an existing browser
Sign in to the target website yourself in the intended Tabbit profile. Confirm the current account, workspace, and page before the agent reads anything. Skill discovery, local launcher reachability, and site login are separate gates. A Reddit user reported that Muse's own browser was blocked on sites they needed; that is one user's report, not a general compatibility result. Keep a fallback to a local browser agent. See the session reuse guide and CLI versus MCP guide.
Ask for a read-only dashboard brief
Use a page with a few visible rows and stable IDs. The model should collect evidence and draft a note, not change the underlying records.
Use the Tabbit skill in this local agent to inspect the currently open [approved dashboard].
Read only the visible rows for [IDs]. Do not open attachments or hidden fields.
Do not edit, save, assign, submit, or send anything.
For each row, return the exact ID, visible status, title, and source URL. Mark missing values "not visible".
Prepare a browser draft titled "Muse dashboard review — [date]" with the table and unresolved items.
Stop before saving or sending. Show the full draft and source links for my review.Expected output is a short table with one source link per row. Open each source in Tabbit and verify the status and title. If the selected Muse route is Muse Wonder Full in Muse Code, it cannot directly follow this prompt through Tabbit CLI unless your local setup has a separately verified handoff. Copy the task to a local Codex or Claude Code agent that can discover the Tabbit skill and run the launcher. Do not imply that the API and Muse Code have identical tools. First-task checklist
Keep the decision and the write separate
| Check | Pass evidence | Stop if |
|---|---|---|
| Model identity | muse-wonder-1.0 appears in the API client's selector, or Full is visible in Muse Code. | You only see a generic “Wonder” label with no route details. |
| Client capability | The local client has the Tabbit skill and can execute the installed launcher. | The skill loads but shell execution is remote or unavailable. |
| Browser session | The intended user, workspace, and record page are visible in Tabbit. | Login expired, site is blocked, or workspace differs. |
| Draft review | Every claim links to a visible row and you approve each proposed edit. | The draft infers values or asks to save/send without review. |
Muse Wonder can be a fit when you have API access and a local coding agent that can run the browser tool. Muse Wonder Full may be useful inside Muse Code, but its availability does not establish a Tabbit CLI connection. For a broader overview of browser control, read how agents control browsers, profiles and tab groups, and the complete CLI guide. Keep the first run bounded and stop at the named draft.
FAQ
Are Muse Wonder and Muse Wonder Full the same model route?
No. Muse Wonder uses API model ID `muse-wonder-1.0`; Muse Wonder Full is available only in Muse Code on eligible subscriptions.
Does setting reasoning to max select Muse Wonder Full?
No. Meta says `max` is a reasoning setting for `muse-wonder-1.0` and does not select Muse Wonder Full.
Can Muse Wonder use a signed-in Tabbit browser?
Muse Wonder can be paired with a local client only if that client can select the API model and run the installed Tabbit CLI. Verify the route, skill, launcher, and website login separately.
Can Muse Wonder Full call Tabbit CLI from Muse Code?
That connection is not established by the public availability docs. Unless you verify a local bridge, hand the reviewed task to a separate local agent.
What should the first browser task do?
Read a few visible rows, return each source URL, and stop at a named draft for your review.