agentrun
Workflow DSL for existing agents that uses Jev for the steps that are decisions and calls an agent only when investigation is needed.
Overview
AgentRun (agent.run()) is a workflow language for the agents you already run. You define repeatable steps, use Jev for focused decisions, and call an agent only when work needs investigation. Your application keeps its tools, model access, permissions and budgets.
The support example routes requests: resetting a password or finding an invoice takes one decision call and no agent call, while investigating a failed payment takes one agent call and two decision calls, and an unresolved payment is escalated for review. Demos use scripted tools, Jev answers and agent responses, so they need no API key. There is also a Pi extension to build workflows with your agent.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What kind of support request is this and which path should it take?(inferred) | help answer · investigate · escalate |
| Yes / no | Is the payment issue resolved after investigation?(inferred) | yes · no |
How it works
- Define a workflow in the DSL
- Decision steps call Jev for a typed answer
- Branch: answer directly, call an agent, or escalate
- Report agent and decision call counts per case
Strengths
- Avoids agent calls for simple cases
- Demos run without keys
- Fits existing tools and budgets
Caveats
- Beta package
- Live use needs TypeSafe and agent credentials
Getting started
npm install @parcha/agentrun-dsl@beta && npx agentrun demoAt a glance
- Type
- Library
- Status
- beta
- Language
- TypeScript
- Runtime
- Node library/CLI, Pi extension
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
A workflow DSL for agents with a Jev package for the steps that are decisions rather than generation.
MrJev/awesome-jev
Frequently asked questions
When is Jev used versus an agent?
Jev answers the decision steps; an agent is called only when investigation is needed.
Can I try it without keys?
Yes, the demos use scripted decisions and responses.
Is there editor support?
A Pi extension helps build workflows with your agent.
Related projects
n8n Jev Classification node that branches items by category with a Needs Review branch, plus parallel and batched requests.
ProjectbetaTypeScriptOpen-source workspace for building, deploying and monitoring AI agents and workflows, listed as integrating TypeSafe System One.
ProjectproductionTypeScriptn8n community node for Jev that adds a typed smart if-statement: choice, score and noul with probabilities in workflows.
ProjectbetaTypeScriptn8n node that sends a state and a map of typed questions to /v1/systemone in one request and returns per-question answers.
ProjectbetaTypeScriptn8n community node for TypeSafe's System One API with yes/no, choice and score questions, retry handling and an AI Agent tool.
ServicebetaTypeScriptPython library to batch, gate, trace and calibrate thousands of typed Jev decisions, maximising questions per call.
ProjectprototypePython