# Jevpipe

> Updated 2026-09-28 · type: tool · category: ai-infrastructure · status: active · rev 1

Jevpipe is a Unix filter that pipes data into a cheap System-1 model and returns typed decisions — yes/no, routing, or scores — for bulk agent judgments.

- Open source: yes (Apache-2.0)
- Self-hostable: no
- Pricing: usage-based via OpenRouter, ~$0.023 per 1,000 decisions; cost caps via --max-cost
- Best for: An engineer who wants to drop cheap, typed LLM judgments into a shell pipeline — classify, route, or score bulk records with a Unix filter — and cap spend per run, accepting an OpenRouter API key and a hosted model as dependencies.
- Not for: Anyone needing fully offline or self-hosted inference (the client is OSS but the 'Jev' System-1 model is hosted via OpenRouter), or tasks demanding high-stakes accuracy beyond a cheap first-pass classifier.
- Last verified: 2026-09-28

- **Canonical:** https://gtmstacker.com/registry/tool/jevpipe/
- **Source:** [fabianboth · GitHub](https://github.com/fabianboth/jevpipe)
- **Tags:** ai-infrastructure, agent-tooling, classification, unix-filter, cli, llm
- **Repository:** https://github.com/fabianboth/jevpipe

## Is Jevpipe open source?

Yes, Jevpipe is open source under the Apache-2.0 license.

## How much does Jevpipe cost?

usage-based via OpenRouter, ~$0.023 per 1,000 decisions; cost caps via --max-cost (usage).

## Can I self-host Jevpipe?

No, Jevpipe is not self-hostable.


---

Jevpipe is a Unix filter that pipes arbitrary data into a cheap "System-1" model and returns typed decisions — yes/no, multiple-choice routing, or scores — for bulk agent judgments. Open source: yes (Apache-2.0) for the client, but it depends on a hosted model via OpenRouter, so it is not fully self-hostable. Pricing is usage-based (~$0.023 per 1,000 decisions); it launched 2026-09-28 with 1 star.

## What it does

Jevpipe makes an LLM behave like any other command in a shell pipeline. You pipe data into it, and it returns a typed decision — a yes/no, a choice from a set (for routing), or a numeric score — by sending the input to a cheap, fast "System-1" model called Jev, hosted through OpenRouter. Because it is a Unix filter, it composes with `grep`, `xargs`, and the rest, which makes it a natural fit for bulk judgments: classify thousands of records, route items, or score candidates in one pass. A `--max-cost` flag caps spend per run. Open source: yes (Apache-2.0) client; hosted-model dependency; usage-priced.

## Provenance

- Apache-2.0 client per repo; 1 star; launched 2026-09-28; Unix filter piping stdin into a hosted "Jev" System-1 model (needs an OpenRouter API key), returning typed decisions (yes/no, multiple-choice routing, scores); usage-billed via OpenRouter (~$0.023/1,000 decisions) with --max-cost caps (github.com/fabianboth/jevpipe, verified 2026-09-28).
- Not fully self-hostable: the client is OSS, the model is hosted. Represented here as "OSS client, hosted-model dependency," not as a self-hosted tool.
- Maker-reported: 83.2% classification accuracy at $0.023/1,000 decisions; found 1,272 relevant functions across 470 searches in 5 languages — the maker's own numbers (vendor-claim), not independently reproduced.
- Curated from the GTM Stacker signal registry (2026-09-28 pass); license/facts independently verified 2026-09-28.

## Why it matters for a GTM stack

A lot of GTM work is cheap, high-volume judgment: is this lead in-ICP, which segment does this reply belong to, how relevant is this scraped result. Jevpipe turns that into a shell one-liner — pipe the records in, get typed decisions out, cap the cost — which is exactly the shape of a bulk enrichment or triage step. Pricing is usage-based (~$0.023 per 1,000 decisions). The honest read: two caveats matter. It is an OSS client but leans on a hosted System-1 model via OpenRouter, so it is not a self-hosted or offline tool and it bills on your account; and the 83.2% accuracy is the maker's own figure on a day-one project — fine as a cheap first-pass filter, but validate it against your labels before it gates anything consequential.