Autotab

Autotab acts in other systems through their interface, so it can work in ERP, RMS and TMS tools without an API. The only named integration step is Google Sheets, which Autotab reads by downloading the sheet as a CSV, and Autotab describes no connector catalog or custom tool support.

SourceAutotab, autotab.com and docs.autotab.com Google Sheetsread 2026-10-05

A skill is an editable list of steps that mixes recorded deterministic actions with agent steps, loops and conditionals, passing data between steps through the scratchpad. Agent steps take a goal in natural language and work out the actions themselves, and Autotab recommends them only for specific subtasks that are hard to teach step by step.

SourceAutotab, docs.autotab.com overview and Agentread 2026-10-05

Autotab is taught a workflow from a recording, a video message or a prepared document, which becomes the skill's instructions, and each skill holds all the context about its task. Autotab describes no store of the customer's knowledge that the agent searches. A Google Sheet can be downloaded as a CSV and loaded as data for a run.

SourceAutotab, autotab.com, docs.autotab.com overview and Google Sheetsread 2026-10-05

Users review and edit the steps Autotab learned before it runs them and can watch a local run in its browser, which keeps the agent to steps the customer has checked. Autotab describes no approval step during a run before an action goes through. Its docs advise using open ended agent steps only for specific subtasks, so users know what will happen.

SourceAutotab, docs.autotab.com overview and Agentread 2026-10-05

Cloud runs use dedicated isolated VMs, passwords never leave the device and secrets are never logged. Inputs are encrypted and sent straight to the execution environment without passing through Autotab's servers, data is encrypted in transit and at rest, and access uses secure tokens tied to each user account. Autotab describes no roles, SSO or other access model and names no attestation on its site or in its docs; a trust center sits at trust.autotab.com.

SourceAutotab, docs.autotab.com Cloud Security and autotab.comread 2026-10-05

The Run API lists and retrieves each run and returns a video URL of the run, and the app shows the steps a skill executed. Users can also watch a local run live in Autotab's browser. Autotab does not say how long runs and videos are kept or how to export them.

SourceAutotab, docs.autotab.com Video Url, overview and API Quickstartread 2026-10-05

The scratchpad carries data across the steps of a run, and Autotab can use it as context to fill a form, reason about the next step or act on the data. Learned skills persist as instructions, and synced sessions hold browser login state and preferences; Autotab keeps only the most recent browser session. Neither is a store the agent builds up as memory.

SourceAutotab, docs.autotab.com overview, Session and Cloud Securityread 2026-10-05

Skills run either locally on the customer's own computer or in the cloud on dedicated isolated VMs, with inputs and secrets sent straight to the execution environment; no region list is published. Each cloud environment is destroyed after its skill finishes. A computer running local skills while unattended needs automatic sleep turned off while on power.

SourceAutotab, docs.autotab.com Running and Cloud Securityread 2026-10-05

Skills are taught per customer by recording or demonstration. The use cases in the docs (employee onboarding, inventory updates, restaurant booking, lead extraction) are worked examples rather than packaged skills to adopt. Autotab's users are domain experts rather than software engineers, so each team teaches the skills it needs.

SourceAutotab, docs.autotab.com use cases overview and overviewread 2026-10-05

Cloud runs can be put on a schedule, and runs also start manually in the app or through the Run API, with webhooks reporting run events. Scheduled runs may need a fresh login now and then, since some sites log users out after a while.

SourceAutotab, docs.autotab.com Running, Session and Webhooksread 2026-10-05

The models Autotab runs are not named in its docs or on its site, and no model choice is offered. Agent steps work out their own actions from a goal written in natural language, while recorded steps replay a fixed action.

SourceAutotab, docs.autotab.com llms.txt index and Agentread 2026-10-05

A Developer API with an OpenAPI file starts, lists, retrieves and cancels runs and lists skills, with JavaScript (npm autotab) and Python (autotab-ai) clients, API keys and webhooks. Webhooks report when a run finishes, is cancelled or times out, and each carries a signature in a header that the receiver checks with HMAC SHA256 against its webhook secret.

SourceAutotab, docs.autotab.com API Quickstart and Webhooksread 2026-10-05

Users verify and edit the steps Autotab learned and review runs through their steps and recordings. Autotab describes no evaluation harness, scored test cases or quality gate. It says it is built for reliability that does not hallucinate, so a task completes correctly every single time.

SourceAutotab, docs.autotab.com overviewread 2026-10-05

Autotab works in its own secure local browser on the user's computer or on dedicated cloud virtual machines, navigating applications, clicking and filling forms like a person. Agent steps handle pages whose layout varies, such as clicking connect on LinkedIn profiles laid out differently, and synced sessions let cloud runs skip logging in again on most websites.

SourceAutotab, docs.autotab.com Running, Agent and Sessionread 2026-10-05