evoke
Natural-language CLI where Jev picks which installed 'reflex' programs a sentence calls, running them only when sure enough.
Overview
evoke turns a sentence into a call of small programs called reflexes, which anyone can write, share and install from Git. Type one sentence and evoke reads it against every installed reflex, picks the ones it asks for, and fills their inputs from your words or your own lists. It shows the plan before anything runs, runs when it is sure enough for what the program changes, asks when something is missing or unclear, and always asks before anything irreversible.
According to the list summaries, Jev is the classifier that chooses reflexes and inputs, and a user overlay can tighten a reflex's effect but never loosen it. evoke ships a CLI you talk to, a package manager for reflexes and a TypeScript SDK over the same core, which never touches files or the network itself. The README emphasises measured behaviour, tested word for word, and outlines a vision of signed, cross-machine plans where each step runs where its program and approver live.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which installed reflex(es) does this sentence ask for?(inferred) | — |
| Choice | Which value from a closed source fills this reflex input?(inferred) | — |
How it works
- User types one sentence
- Jev matches it to installed reflexes and fills inputs
- evoke shows the plan with probabilities
- Run when confidence suits the effect; ask on gaps or irreversible steps
Strengths
- Effect-aware confidence gating
- Values come from closed sources
- Overlays can only tighten permissions
Caveats
- Requires a TypeSafe key
- Ecosystem of reflexes is small (about fifteen for Mac in the collection)
Getting started
curl -fsSL https://evoke.build/install.sh | sh on macOS or Linux.At a glance
- Type
- CLI
- Status
- beta
- Language
- TypeScript
- Runtime
- CLI + SDK
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Software, by reflex. A sentence becomes a call of a small program, chosen by Jev, TypeSafe AI's classifier, and run only when it is sure enough. Reflexes are recipes anyone can write, share and improve. A CLI you talk to, a package manager for reflexes from git, and a TypeScript SDK.
hellogumbo/awesome-jevTurns a sentence into a call of a small program you installed from Git, run only when the confidence gate allows. A CLI, a package manager for those recipes, and a TypeScript SDK over the same core; your overlay may tighten a reflex's effect but never loosen it.
MrJev/awesome-jev
Frequently asked questions
What is a reflex?
A small program with a readable manifest that evoke can call from a sentence.
When does evoke run without asking?
Only when confidence is high enough for what the program changes; irreversible steps always ask.
How do I install it?
curl -fsSL https://evoke.build/install.sh | sh.
Related projects
DeepSeek Harness plugin with 12 opt-in Jev or Luna Decisions judgments: skill and file selection, supervision, filtering and approval help.
PluginbetaTypeScriptopencode plugin that routes the agent's small multiple-answer decisions to Jev and labels each answer auto or escalate by confidence.
PluginbetaTypeScriptSix-file DSPy coding agent where Jev answers closed questions (safety, tool fit, history, delegation) and Python owns policy.
ProjectprototypePythonSkill and CLI that compares a shortlist of agents, MCP servers or skills against a task and permission ceiling, abstaining on weak leads.
MCP serverprototypeTypeScriptStrict-allowlist shell harness where Jev walks a TOML decision tree to pick a bash command, flags and paths, confirming writes.
ProjectprototypePythonMinimal agent loop where Jev selects each action and approves tool calls, a chat model fills arguments, and MCP tools execute.
MCP serverresearchPython