User repository on GitHub
System One Search: navigate and trace code with TypeSafe judgments and repository evidence
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Tools & CLIs
S1s
Repository navigation tool that pairs a lexical index and reference graph with small typed Jev judgments: Find narrows to candidate files, Map locates a subject's members, and Explain builds an evidence-backed graph.
- Stars
- 0
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- cpaczek
- Last verified
- Oct 2026
About this project
Find the file. Map the subject. Follow the flow. System One Search (s1s) navigates a Git repository using code and Jev, TypeSafe’s decision model. The engine does not use an LLM to generate explanations. Its answers consist of source paths, identifiers, extracted comments and real references. Find narrows a question to candidate files and verifies relevant source lines. Map locates the files that belong to a subject. Explain traces a subject across files and builds a chart with an evidence-backed walkthrough.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- The engine never generates prose, so answers are source paths, identifiers, extracted comments and real references.
- `index` and `check` run entirely locally; only Find, Map and Explain send descriptors and source evidence to TypeSafe.
- The library takes an injected `Client`, so a harness can add budgeting, cancellation, logging or fake responses for tests.
- `check` audits which files your harness actually read and which reference neighbors remain, with no model call.
Watch out
The README states the boundary plainly: a high-confidence result is still a model judgment, and a flow is a bounded static view rather than a runtime trace. Unsupported syntax, dynamic wiring, vocabulary gaps, search caps and unexamined files can all hide relevant behavior. `check` reports which files a harness examined and which reference neighbors remain, and explicitly does not establish semantic completeness. Benchmark results are specific to their questions, revisions and runs, with no universal accuracy or completeness claim, and the optional embedding baselines belong to the benchmark rather than the search engine. Source-installable and MIT licensed, but npm publishing is a later step, so `private: true` is intentional. Requires Node 24+, pnpm and Git, and core tests also need `rg` on PATH for the real grep baseline plus Python 3 for the offline dense-helper tests. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Two structural differences. The engine does not use an LLM to generate explanations, so answers are source paths, identifiers, extracted comments and real references, which the README sums up as recall being code and precision being TypeSafe. Candidate proposal is a lexical index plus a reference graph; small typed judgments only decide which proposed pieces belong. Ranking, caps, graph construction and layout are therefore ordinary code, and every question and threshold lives in `src/questions.ts`.
Find, Map and Explain send selected repository descriptors and source evidence to TypeSafe. Indexing and coverage checks do not. `pnpm s1s index --repo <path>` inspects the local index without calling TypeSafe, and `pnpm s1s check --repo <path> --examined src/router.ts,src/auth.ts --json` audits which files a harness actually reviewed, also with no model call. The demo playground behaves the same way, restoring a completed run with its treemap and trace without another inference request.
It covers what you read, not what you missed. `check` reports which files a harness has examined and which reference neighbors remain, and the README says in the same sentence that it does not establish semantic completeness. The intended workflow is that you pass the paths your harness actually read rather than every search candidate, so a small remaining-neighbor count means your reading covered the graph's edges. It is not evidence that the answer was right.
Source install for now. The project is MIT licensed and `private: true` is intentional because npm publishing is a later step. Requirements are Node.js 24+, pnpm and Git, and the TypeScript runs directly in Node with no compilation step for the library or CLI. You clone, run `pnpm install`, copy `.env.example` to `.env` and set `TYPESAFE_API_KEY`. `--repo` defaults to the working directory, must be a Git checkout, and the index starts from tracked files read at their current working-tree contents.
Every question and threshold lives in src/questions.ts, and your data goes in the request state rather than into question text. Ranking, caps, graph construction and graph layout are ordinary code, so the whole decision surface is readable without touching the model path. The README asks you to measure any strategy change on public questions with a fake client before extending it.
find and check return JSON with --json, explain writes a self-contained evidence-backed chart with --out and bounds the walk with --depth, and find takes --scope to keep the search inside a subtree such as src. serve also puts an advanced playground on http://localhost:4747/playground, and a public demo with pre-indexed repositories sits at s1s.iar.dev.
A flow is a bounded static view, not a runtime trace. The README lists unsupported syntax, dynamic wiring, vocabulary gaps, search caps and unexamined files as things that can hide relevant behavior, and says plainly that a high-confidence result is still a model judgment. docs/limitations.md is the place to read before letting a result guide a change.
More Jev resources like this
38GitHub stars
Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
samdotmak/jev-recallTypeScript
samdotmak/jev-recallTypeScript36GitHub stars
Stock decision terminal that gathers Valyu market evidence across five categories and uses Jev for a typed 30-day outlook with probability, signals, and timing.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
unicodeveloper/jevocksTypeScript
unicodeveloper/jevocksTypeScript4GitHub stars
Playful chat demo that makes Jev hold a conversation one word at a time by picking from a compact 979-word vocabulary through repeated Choice calls.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
mkotlikov/jev-grugTypeScript
mkotlikov/jev-grugTypeScript4GitHub stars
Shell CLI that sends state plus noul, choice, and score questions to Jev in one request and prints probabilities, choices, and rubric positions for pipes and scripts.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
y0usaf/typesafe-cliTypeScript
y0usaf/typesafe-cliTypeScript5GitHub stars
Local-first app that extracts document text on the machine and files it into configured category folders using typed Jev decisions for category, confidentiality, prompt-injection risk and subject, with a per-run audit that previews and undoes every move.
Jev Tools & CLIs#community#jevbest#catalog-import
7GitHub stars
PDF search that asks Jev one noul per line, does this line answer the query, and ranks the lines by the returned probability, so matches highlight without embeddings, a vector index or a chunking strategy. Text is extracted locally with pdf.js and the key stays on the server.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
kylemclaren/jevpdfTypeScript
kylemclaren/jevpdfTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.
Jev Tools & CLIs#gateway#vercel#cloudflare
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.
Jev Tools & CLIs#classification#email#security
Read article
From the Jev community
Posts from builders shipping with Jev right now.