WorkerKitJoin the waitlistAI workers for TypeSafe
Ready-made TypeSafe AI workers bring System One decision models into inbox, calendar, spreadsheet, document, code, and web jobs, with typed outputs, confidence-aware routing, and per-worker permissions.
- Sheets Lead Qualification Decision model Returns lead-fit classifications from spreadsheet rows against your ideal customer profile.
- Internal Request Router Decision model Labels incoming requests for IT, HR or Finance and escalates requests needing private handling.
- Meeting Prep Needed Screen Decision model Reads the week ahead and says, meeting by meeting, whether the owner needs to prepare and what kind of preparation it is.
- Critical Email Finder Decision model Reads a window of inbox mail and returns only the messages where something breaks, money is lost, a person is blocked or a deadline passes if the owner does not see them today.
- Web Research Relevance Filter Decision model Finds and reads web pages on your research topic and returns relevant sources with typed judgments.
- Release Note Classifier Decision model Returns release-note sections and merge judgments for pull requests so an agent can assemble a changelog.
- Meeting Entity Relevance Finder Decision model Finds matching meeting that was actually about a customer, project, vendor, person or topic you name, including the internal ones the target was never invited to.
- Brand Mention Urgency Radar Decision model Finds public brand mentions that merit a company response or investigation.
- Email Jev Classifier Decision model Sorts a window of inbox mail into your own categories and returns one typed row per message, so an agent can read the shape of an inbox in one call.
- Critical Meeting Finder Decision model Reads the week ahead and returns only the meetings where revenue, a decision, a customer or an external commitment suffers if the owner skips them or walks in unprepared.
- Email Reply Needed Screen Decision model Reads a window of mail and says, message by message, whether the owner owes a reply and who owes the next move.
- Sheets Jev Classifier Decision model Classifies spreadsheet rows into your own categories and returns typed results for the calling agent.
- Calendar Jev Classifier Decision model Sorts the week ahead into your own meeting categories and returns one typed row per event, so an agent can read the shape of a calendar in one call.
- GitHub Issue Triage Decision model Reads a repository's open issues and decides what each one is, how badly it hurts, and whether it needs a maintainer.
- Email Entity Relevance Finder Decision model Finds matching message that is actually about a customer, project, vendor, person or topic you name, internal and external, and names the false positives a keyword search returns silently.
- Document Content Relevance Decision model Finds documents whose readable content addresses your chosen topic or business requirement.
- Sheets Feedback Router Decision model Returns a recommended handling category for each customer-feedback row in a spreadsheet.
- Inbox Reply Labeler Decision model Labels requests needing a reply, urgent requests and bulk mail so your inbox is ready to work.
What is a TypeSafe AI worker?
A TypeSafe AI worker is a WorkerKit worker built around TypeSafe's machine-consumable decision primitives. Current kits use Jev to evaluate independent Choice, Score, and Noul questions over bounded source state, then combine those answers with routing logic in the worker. TypeSafe supplies the decision model; WorkerKit controls data access, execution, actions, schedules, and receipts.
What these workers do
Choose a queue for customer feedback
Uses a Choice question over mapped Google Sheets feedback to return incident, support, product, praise, noise, or needs-context.
Score the consequence of an email
Uses a Score rubric alongside a criticality question to surface messages where delay could block work, lose money, or miss a deadline.
Test whether a meeting needs preparation
Uses a Noul question and a preparation-type choice over calendar evidence, then returns meetings that cross the review threshold.
Combine relevance signals for research
Judges page or document relevance with separate typed questions, then applies code-defined thresholds and returns the matching source references.
Route classifications into app actions
Maps typed outcomes to explicit routes, such as adding an inbox label or moving a message, only when the installed worker grants that action.
How a TypeSafe AI worker runs
What TypeSafe handles
What WorkerKit handles
How you stay in control
TypeSafe questions, answered.
What is TypeSafe AI used for in WorkerKit?
TypeSafe provides the structured decision model used by this category of AI workers. WorkerKit sends bounded source state to Jev as typed Choice, Score, and Noul questions, then uses the answers and confidence in explicit routing logic. This suits classification, screening, scoring, and routing jobs whose outputs must be consumed by software.
Are TypeSafe and Jev the same thing?
No. TypeSafe is the company and model platform, while Jev is its flagship System One model. The TypeSafe category groups workers built with that decision-model approach. The Jev category is the model-specific view of the same current pack, so visitors can find it by either the company or model name.
Why use typed questions instead of asking for JSON?
Typed questions define the allowed answer shape before the model runs. Jev returns structured Choice, Score, or Noul results and confidence directly, while WorkerKit validates the decision specification and owns the routing code. That removes prose parsing from the decision path, although source quality and model uncertainty still require sensible thresholds.
Can TypeSafe AI workers act across my apps?
Yes, through WorkerKit's app tools and only within the permissions attached to that installed worker. The decision model does not receive general account access. WorkerKit supplies the bounded state and can apply a declared route, while each run receipt records the result, source reference, and action outcome.
Can a TypeSafe decision still be wrong?
Yes. A typed output is still a model judgment, and its quality depends on the available state and criteria. Jev includes confidence information, and WorkerKit lets a kit set a floor that escalates uncertain items. Source coverage, omissions, and warnings should be checked before treating a batch as complete.