GLM-5-Turbo can make tool calls, but that does not sign it in to your websites or connect it to a local browser. For a browser task, the model provider, agent client, Tabbit CLI and selected browser profile each need to work. This guide uses a project handoff memo as an example: read approved team pages, keep a source ledger and save a named draft for a teammate to review.
Confirm the model and client separately
Z.AI currently documents GLM-5-Turbo under the direct API ID glm-5-turbo, with function calling and an OpenAI-compatible chat-completions endpoint. The catalog ID zhipuai/glm-5-turbo is not the direct request string in that guide. A gateway may require its own prefix, and a model listing does not guarantee that your chosen agent client supports the model or local tools. Check the live provider catalog and account access before changing browser settings. See the official Z.AI model guide for the current endpoint and request shape, and the Z.AI model directory for its current product listing. Neither source establishes that your local agent client has configured credentials or can execute browser tools.
Verify the route before opening team pages
| Layer | Verify | If it fails |
|---|---|---|
| Provider | Current Z.AI docs and account list glm-5-turbo | Check endpoint, region and entitlement with the provider |
| Agent client | It selects the provider/model and supports function calls | Confirm the client’s model ID and tool-loop setup |
| Local tool | The client can run Tabbit CLI and load its skill | Check launcher path and local command permission |
| Browser session | Intended profile shows the approved project page | Sign in manually; verify account and domain |
The handoff case starts with an explicit page list. Sign in manually to the approved site in the selected browser profile, then confirm the visible account, domain and page title. A successful model request is not proof that the browser is on the right account. A task name helps distinguish your work, while the website session still comes from the selected profile.
Keep the handoff evidence-based
| Source type | Capture | Do not infer |
|---|---|---|
| Project brief | Visible title, updated date and stated scope | A date or goal absent from the page |
| Decision log | Decision text, owner if shown, and source URL | Approval from an informal mention |
| Release status | Exact status wording and visible date | That “ready” means published |
| Open question | Unresolved text and the page that contains it | An answer from a model guess |
Suppose a team member needs a short project handoff from three approved pages: a project brief, a decision log and a release status page. The deliverable is a shared document called Project handoff — source-checked draft. It is a draft for review, not an instruction to share or publish.
Try a draft-only handoff
Install Tabbit Browser and sign in to the approved workspace yourself. Invoke the installed Tabbit skill through your local client. Pi uses /skill:tabbit. Open a separate task tab so you can keep working in your other tabs.
Use the saved-draft step only in a document workspace where you have explicitly approved creating a draft. Otherwise, return the named memo in the agent response.
Use the selected local Tabbit task named “GLM project brief draft”. Open only the three URLs I provide: the project brief, decision log and release status page. Read the visible title, date, status and relevant facts. Create a document draft named “Project handoff — source-checked draft”. Include a source ledger with one URL per claim, distinguish confirmed facts from open questions, and quote exact wording when status is unclear. Do not publish, send, share, submit a form or change account settings. If login or MFA appears, stop and ask me to sign in. If the wrong account, tab or domain is visible, stop. After saving, report the document title, URL and visible Draft status, then wait for review.Keep the result in Draft and inspect every source URL. If a page is unavailable, contradictory or asks for login, record the gap and stop; do not fill in missing facts from memory. Community discussions also show why narrow scope matters: commenters describe limiting the browser profile to a few needed sites, checking permissions and retaining human approval before consequential actions. These are community preferences, not proof of any product’s security model.
For setup and follow-up, see complete CLI guide connect to an existing browser first-task checklist browser MCP versus CLI research with source links CMS draft workflow let an agent control a browser export browser data to CSV Claude Code browser permissions agent-browser CLI tutorial. The links cover local CLI setup, browser selection, research receipts, permissions and draft workflows; they are related steps, not evidence that a particular GLM client integration has already been tested.
The useful division is simple: GLM-5-Turbo supplies the model; the agent client connects its tool calls to local Tabbit CLI; the selected browser profile supplies the site session. Verify each layer, keep the allowed page list narrow and require a named draft with source links. The workflow here is an example; the only session check available for this assignment was a Tabbit fixture, not a team document or production site.
Sources
Z.AI GLM-5-Turbo API guide: direct model ID, endpoint and function calling; checked 2026-11-10.
Z.AI AutoClaw model directory: current official listing; checked 2026-11-10.
Tabbit CLI complete guide: local task workflow.
FAQ
What is GLM-5-Turbo’s direct Z.AI API model ID?
The Z.AI guide uses `glm-5-turbo`. `zhipuai/glm-5-turbo` is the catalog identifier. Other gateways may require different prefixes.
Does GLM-5-Turbo support tool calls?
Z.AI documents function calling for this model. Your client still needs to support the provider route and connect calls to a local tool such as Tabbit CLI.
Can GLM-5-Turbo use my signed-in browser?
Only when the chosen client can run local Tabbit tools and reaches the browser profile where you signed in. The API model page alone does not establish that integration.
Will the agent sign in for me?
Sign in yourself in the selected browser profile. Stop if the page asks for a password, MFA or another account action.
Does a new Tabbit task preserve my site session?
A task separates its task state; site login depends on the selected browser profile and the site session. Check the visible URL, account and page title before proceeding.
Can it publish the handoff automatically?
This example stops at a saved Draft state. Review the content and sources yourself before sharing or publishing.