DataJev
September 21, 2026 · View on GitHub
All projects · Developer tools
System-1 control loop for data analysis agents: an LLM proposes analytical steps, Python executes them, and TypeSafe Jev chooses the next trajectory action (CONTINUE / SWITCH / VERIFY / STOP). A heuristic controller supports offline smoke tests without keys.
| At a glance | Details |
|---|---|
| Source | Source |
| Maintainer | zzz1YAO. Independently curated; this page is not an upstream submission or endorsement. |
| Format | Python package datajev 0.1.0 (CLI + controllers; uv managed). |
| Requirements | Python 3.12; uv; for live control set TYPESAFE_API_KEY and an OpenAI-compatible analyst (DATAJEV_LLM_*); offline --controller heuristic / examples/quickstart.py --offline need no keys. |
| License | MIT. |
| Disclosure | AI-assisted catalog review; no affiliation. Listing is not an endorsement. Offline pytest 48 passed; examples/quickstart.py --offline OK. Live TypeSafe/LLM smoke not run. |
When to use
Use it when you want trajectory control separated from generative analysis on tabular data. Prefer plain notebooks when a single scripted path is enough. Heuristic/offline mode does not exercise Jev.
How it works
datajev/controllers/jev.py posts compressed analytical state (goal, schema, recent results—not the raw CSV by default) to https://api.typesafe.ai/v1/systemone. Python applies the chosen verb and continues the loop. Analyst prompts and tool traffic use the configured LLM endpoint separately.
Get started
git clone https://github.com/zzz1YAO/DataJev.git
cd DataJev
git checkout b42d67c7f2034a802451615c7e60e2c39c174fc3
uv sync
uv run pytest -q
uv run python examples/quickstart.py --offline
For live runs, copy .env.example, set TypeSafe and LLM keys, then uv run datajev analyze … (billed by both providers).
Examples and demos
examples/quickstart.py --offlineand sample CSVs underexamples/.- This listing: pytest 48 passed; offline quickstart OK. No live TypeSafe call.
Limits and data handling
Jev sees a structured analytical state derived from the run; the analyst LLM may see richer intermediate artifacts depending on configuration. Heuristic control is not a neural judgment. Quality claims were not independently measured.
Review and maintenance
Reviewed on 2026-09-21 at commit b42d67c: MIT; AI-assisted source review of README, LICENSE, datajev/, pytest, and offline quickstart. No live TypeSafe call.
Related: daf-jev, decision-first, hunch.