Jev + Browser Use
Start with the browser-agent pattern: state and allowed actions go to Jev, then browser code executes the selected action.
Explore the browser pattern →How to use Jev, from first click to first API call.
Start in the Playground without writing code. When the decision works, create a TypeSafe API key and connect the hosted Jev model through HTTP, Python, JavaScript, or the official TypeSafe skill for coding agents.
Official docs checkedOfficial skill commandsNo API keys shown or collected
JUST EXPLORINGPlaygroundNo local install
orBUILDING WITH AN AGENTOfficial TypeSafe SkillCodex · Claude Code · compatible agents
+CALLING JEVAPI key + SDKPython · JavaScript · HTTP
01
FASTEST TEST · NO LOCAL INSTALL
Try one bounded judgment in the Playground
The official Playground is the quickest way to see whether a task fits Jev. Paste a representative state, add one narrow question, and inspect the typed answer and probabilities before building an integration.
- Open the TypeSafe Console Playground and log in.
- Paste a short state, such as a customer message or current agent state.
- Add a Noul, Choice, or Score question.
- Try normal, ambiguous, and adversarial examples. Decide what low confidence should do.
02 · TYPESAFE CONSOLE
Create an API key without exposing it
You only need this step when code will call Jev. The interface may change, but the security rule does not: copy the new key once, store it outside your source code, and never publish it.
1Open the TypeSafe Console
Log in, then select API Keys from the sidebar. The Playground remains available for no-code tests.
2Start a new key
On the API Keys page, choose Create key.
3Name and create it
Use a name that identifies the project or environment, then create the key.
4Copy the new key when it appears
The console shows the full value once. This tutorial image uses a redacted example; do not publish an unredacted key.
Store it outside screenshots, chat, and source code
Open API Keys ↗Save the value as TYPESAFE_API_KEY in a local or deployment environment variable. If a real key appears in public material, rotate it.
03
OFFICIAL AGENT SKILL
Let Codex, Claude Code, or WorkBuddy help with setup
TypeSafe's official skill gives a coding agent current guidance for designing questions, reading the live documentation, and composing Jev workflows. Use one installation method for your environment.
Official copy-to-agent setup prompt
Install the TypeSafe skill. If you're in Claude Code, run `claude plugin marketplace add typesafe-ai/skills`, then `claude plugin install typesafe@typesafe-ai`. If you're in another agent, run `npx skills add typesafe-ai/skills --skill typesafe-ai` and select your agent. Use one installation method. You can read the skill directly at https://github.com/typesafe-ai/skills/blob/main/skills/typesafe-ai/SKILL.md (raw: https://raw.githubusercontent.com/typesafe-ai/skills/main/skills/typesafe-ai/SKILL.md). Then use the TypeSafe skill when working on this project.CLAUDE CODE
Official Claude Code commands
claude plugin marketplace add typesafe-ai/skills
claude plugin install typesafe@typesafe-aiCODEX / WORKBUDDY / COMPATIBLE AGENT
Official skills installer
npx skills add typesafe-ai/skills --skill typesafe-aiOne prompt to configure a real project safely
This expanded prompt adds project inspection, API-key handling, one smoke test, and verification to the official installation step.
Copy complete project setup prompt
Set up TypeSafe and Jev for this project.
1. Install the official TypeSafe skill using exactly one supported method. For Claude Code, use the TypeSafe marketplace plugin. For Codex, WorkBuddy, or another compatible agent, use the skills installer and select the current agent.
2. Read the live TypeSafe documentation before writing integration code. Treat it as the source of truth.
3. Inspect this project and choose the smallest server-side Jev example that fits its existing language and framework. Do not add a database, UI framework, or unrelated infrastructure.
4. If TYPESAFE_API_KEY is missing, stop and direct me to https://console.typesafe.ai/keys. Never ask me to paste the key into chat. Add only a placeholder to .env.example, keep the real .env file ignored, and never expose the key in browser code.
5. Implement one smoke test using jev-latest with a short state and one bounded Choice, Score, or Noul question. Keep actions, permissions, thresholds, and fallback behavior in code.
6. Run the relevant typecheck, test, and build commands. Then report the files changed, the command to run the example, and anything I still need to do manually.04
HTTP / PYTHON / JAVASCRIPT
Run the first Jev request from your own program
Jev remains hosted by TypeSafe. Your server sends state and typed questions to POST /v1/systemone; your code receives structured answers and decides whether to act, stop, or ask a human.
Store the key for this shell session
export TYPESAFE_API_KEY='your-key-here'Minimal HTTP request
curl -X POST https://api.typesafe.ai/v1/systemone \
-H "Authorization: Bearer $TYPESAFE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"state": "The customer says their payment failed twice and they need help today.",
"model": "jev-latest",
"questions": {
"is_urgent": {
"type": "noul",
"instructions": "Does this message convey urgency?"
}
}
}'Python
Install
pip install typesafe-sdkMinimal Python example
from typesafe_sdk import Noul, TypeSafeClient
client = TypeSafeClient()
response = client.system_one(
state="The customer says their payment failed twice and needs help today.",
questions={
"is_urgent": Noul(
instructions="Does this message convey urgency?"
)
},
)
print(response.answers["is_urgent"].noul)JavaScript / TypeScript
Install
npm install @typesafe-ai/sdkMinimal JavaScript example
import { choice, TypeSafeClient } from "@typesafe-ai/sdk";
const client = new TypeSafeClient();
const response = await client.systemOne({
state: { message: "The customer says their payment failed twice." },
questions: {
route: choice("Which team should handle this?", {
billing: null,
technical: null,
other: null,
}),
},
});
console.log(response.answers.route.choice);Keep the key server-side. Never place
TYPESAFE_API_KEY in browser JavaScript, a screenshot, a chat message, or a Git commit.Typed output is an interface, not permission to act.
Define choices.
Do not ask Jev to invent an action your code cannot inspect.
Set review behavior.
Uncertainty must lead to a safe fallback.
Protect secrets.
Keys stay in server-side environment variables.
Test your data.
Vendor demos and community reports are not your result.
MORE SETUP PATHS
Find your integration starting point
Dedicated guides for these environments are planned. These entries lead to the existing patterns, connector record, and API instructions.
Jev + MCP
The official skill guides development. A runtime connector is a separate integration; review its source and requirements before using it.
Inspect the community MCP connector →Jev + LangGraph
A dedicated LangGraph guide is coming later. Begin with the hosted API and identify one bounded decision in your graph.
Start with the API →COMMON SETUP QUESTIONS
Before you hand the setup to an agent
Do I need an API key to use the Playground?
The official Quick Start sends users to the TypeSafe Console Playground and asks them to log in. Creating an API key is required when your own code or an agent runtime calls the hosted API.
Does installing the TypeSafe skill give Codex or Claude Code direct access to Jev?
No. The official skill teaches the coding agent how to design and implement TypeSafe workflows. A live call still needs a server-side API key and SDK or HTTP integration, or a separately reviewed runtime connector.
Should I paste my TypeSafe API key into an AI chat?
No. Store it in a local or deployment environment variable named TYPESAFE_API_KEY. Use placeholders in examples and rotate the key if it is exposed.
SOURCES & FURTHER READING
Use official documentation for implementation.
TypeSafe's live documentation and official skill are the source of truth for setup and current API behavior. The WayToAGI article is useful Chinese-language community context, but its claims should be checked against the linked primary sources.