Article
CommunityJev Integrations & Adapters
Mastra Classifiers
Mastra's classifier primitive defines choice, score, and boolean questions against an evaluation model such as jev-latest, returning a probability for each criteria key so workflow branch logic can act on the answer.
- Format
- Article
- Reading time
- 2 min
- Published by
- Community
- Author
- Paul Scanlon
- Last verified
- Oct 2026
- Languages
- TypeScript
- Capabilities
- Requires API key
Highlights
- npm install @mastra/core @ai-sdk/typesafe-ai; classifiers need @mastra/core 1.69.0 or later, added in PR #24458.
- A classifier has an id, a model, and questions typed choice, score, or boolean; boolean answers with the probability the answer is true.
- Questions are keyed by name and all evaluate the same input; instructions and criteria accept any JSON-compatible value, objects and arrays included.
- Each criteria key receives a probability and the highest one becomes the choice the workflow branches on to select an agent.
- Classifiers register on the Mastra instance beside agents, tools, and workflows, and a workflow calls .classifier() before mapping the answer and branching.
Watch out
A vendor feature announcement: it shows one marketing example and links the classifier, workflow, ClassifierProcessor, and createClassifierScorer references, but reports no accuracy, cost, or latency figures. Using it requires a TypeSafe API key, Mastra 1.69.0 or newer, and inputs that fit the declared question types.
Questions about this entry
Short answers drawn from the linked project or page.
A way to use an evaluation model with a defined set of questions: you provide instructions describing what to decide and criteria outlining what can be answered, and a choice returns one of the criteria keys, a score a position from an ordered list, and a boolean a true probability.
Install @mastra/core 1.69.0 or later plus @ai-sdk/typesafe-ai, then create a Classifier with an id, a model from createTypeSafeAi with your API key and .evaluationModel(jev-latest), and one or more questions keyed by name.
Register the classifier on the Mastra instance beside agents, tools, and workflows, then call .classifier(marketingClassifier), map the returned answers.route.choice and the original message, and branch with createStep per agent choice before .commit().
The chosen key and a probability for every criteria key, so the answer shows why the decision was made. Each criteria key gets a probability and the highest one becomes the choice; the example returned pmmLaunchAgent with probability 1.
Yes. instructions and criteria are not limited to strings; both accept any JSON-compatible value, so a criteria description can be an object or an array, and a classifier can hold several questions of mixed types evaluated against the same input.
Paul Scanlon, a technical product marketing manager at Mastra, wrote the 4-minute post on September 28, 2026. It says classifiers shipped after the rise of Jev and links a workshop video, Building a Classifier in Mastra with Jev.
It gives no licence, pricing, rate limits, or accuracy data, and no latency numbers for classifier calls. It points to the Classifier, Workflow.classifier, ClassifierProcessor, and createClassifierScorer references for full configuration options.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
X: OpenRouter Decision Model Rankings
X: herdr + jev, an agent router
X: A group chat bot with decision room routing
More Jev resources like this
An official Mastra template that checks pricing, changelog, and documentation pages against saved captures, has Jev classify each change with its evidence, and routes it to ignore, record, request review, or alert.
Jev Integrations & AdaptersDocs#community#typescript#mastra
Community
Merged PR in the Kilo Code cloud monorepo that replaces its LLM auto-routing classifier with one OpenRouter /systemone Choice request to typesafe/jev-1.13; the PR reports 0.93-0.94 route accuracy on 72 golden cases at roughly $0.00028 per call, with a 250 ms transport-error retry.
Jev Integrations & Adapters#community#routing#coding-agents
Community
Kilo-Org/cloudTypeScript
Kilo-Org/cloudTypeScriptA Claude Code plugin that asks Jev, TypeSafe's System One model, which model tier each prompt needs — haiku, sonnet or opus — and switches automatically, and routes classification, scoring and yes/no judgments to Jev via scripts/jev.py.
Jev Integrations & Adapters#community#github#claude-code
Community
ricky94511/jevTypeScript
ricky94511/jevTypeScriptTanStack AI's TypeSafe adapter covers evaluate only and talks to TypeSafe over fetch, wiring jev-latest into decide() with choice, score, and boolean question helpers and a required TYPESAFE_API_KEY.
Jev Integrations & AdaptersDocs#community#tanstack#typescript
Community
Open-source TypeScript agentic operating system: a core runtime, app for web and mobile, CLI, cloud services and first-party plugins. The directory's curated summary reports it integrates TypeSafe System One decision services for sub-100ms intent classification, action dispatching and confidence-gated tool execution.
Jev Integrations & Adapters#community#agents#framework
Community
elizaOS/elizaTypeScript
elizaOS/elizaTypeScriptOpen-source AI email assistant (Next.js, Postgres, 12k+ stars) with a decision-model layer that can run on openrouter:typesafe/jev-1.13: Jev-backed questions cover rule selection, cold-email detection, thread status, recurring patterns, unsubscribe pages, and sender categorization, each with evals.
Jev Integrations & Adapters#community#email#gmail
Community
elie222/inbox-zeroTypeScript
elie222/inbox-zeroTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.