What exactly does the decision model do in this stack?

As far as the documentation I read goes, it is optional and does typed query classification: intent, service and severity, with a confidence. How that classification then changes search or the generated answer is not spelled out there, so check the project's own configuration before designing around it.

Can I keep my logs off third-party services?

You can, at least for the model. The README offers a locally installed Ollama as the alternative to a hosted LLM key, and your logs land in your own ClickHouse and Qdrant containers. What leaves the machine depends on which provider you configure.

What does it take to get a first answer out of it?

Clone the repo, run setup for the hosted provider or point the provider variable at Ollama, bring up docker compose, then open the dashboard on port 3001. The compose demo profile starts with simulated logs if you would rather not ingest real files yet.

Will it parse a log format it does not list?

The documented formats are JSON, nginx, Apache, syslog and Proxmox, and the README claims unrecognised formats are worked out automatically. That claim carries no accuracy figures, so test the ingest path against a real file before you rely on it.

What components does the stack run?

A Rust backend split into API, parser, retrieval, worker, anomaly and CLI crates, with ClickHouse for storage and Qdrant for vector search, a Next.js dashboard in front, and Groq or a local Ollama for the language model and retrieval steps. All of it comes up through docker compose, and Jev sits in as an optional query-classification step.

What triggers an anomaly alert?

Three patterns, per the README: error-rate spikes around five times normal, new error patterns appearing, and a service going quiet through a volume drop. Alerts go to Slack so you hear about the condition before users report it. The README does not give the detection window or how the five-times baseline is computed.

How do logs get into it?

Three ways. Application code can POST to the local logs endpoint, the CLI can ingest existing files with a format flag and a service name, or a Docker logging driver can forward container output. A separate demo profile starts the stack with simulated logs if you would rather not ingest real files first.

CommunityJev Tools & CLIs

Stratum

Docker Compose log intelligence stack: Rust ingest and retrieval crates over ClickHouse and Qdrant with a Next.js dashboard, where the decision model is an optional query classifier.

See highlights

Open github.com

Stars: 51
Language: Rust
License: APACHE-2.0
Published by: Community
Author: YEDASAVG
Last verified: Oct 2026

YEDASAVG/Stratum

User repository on GitHub

View on GitHub

AI-powered Log Intelligence System - Semantic search, anomaly detection, and root cause analysis for logs using RAG techniques. Built with Rust, Axum, NATS, ClickHouse, and Qdrant.

Forks: 2
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

🧠 Ask your logs questions. Get answers in plain English. Setup • Features • How It Works • FAQ. You know how you have thousands of log files and when something breaks, you spend hours searching through them? Either: Free Groq API key (takes 30 seconds) Or: Ollama installed locally (fully offline, no API key)

Quoted from the project's README. Read the full README.

Tagscommunity jevbest

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Rust

Highlights

The decision model appears as an optional query-understanding step classifying intent, service and severity with a confidence.
Rust crates split into API, parsers, retrieval, worker, anomaly detection and a logai CLI, with a Next.js dashboard in front.
Parsers cover JSON, nginx, Apache, syslog and Proxmox; the CLI ingests files and an HTTP endpoint takes logs from apps.
Anomaly detection watches for error-rate spikes, new error patterns and services going quiet, with Slack alerts.

Watch out

The decision model is optional, and the documentation I read describes it only as typed query classification returning intent, service, severity and a confidence. How those labels steer retrieval or ranking, which thresholds apply, what a call costs and how a key is configured are not documented in that section, so read the project's own configuration before relying on it. The worked answer and the claim that unlisted formats are recognised automatically are demonstration output, not measured accuracy. Setup needs Docker plus either a hosted LLM key or a local Ollama. 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.

More Jev resources like this

claude-code-trace

376

Claude Code Trace renders the JSONL session files in ~/.claude/projects/ as browsable conversations with expandable tool calls, MCP tool names, and token counts, with an optional Jev pass that scores agent efficiency.

Jev Tools & CLIs#community#jevbest

Community

delexw/claude-code-traceRust

james_library

R.A.I.N. Lab is a local-first research environment. Four constrained perspectives investigate a question, quoted spans are checked against a local paper corpus, and experiments are pre-registered with host code, not a model, deciding pass, fail or inconclusive.

Jev Tools & CLIs#community#jevbest

Community

topherchris420/james_libraryRust

doc-router

Rust CLI that checks each PDF page for a usable text layer, extracts text pages locally, sends only the rest to an OCR model via LiteLLM, and merges a page-ordered result.

Jev Tools & CLIs#community#jevbest#catalog-import

Community

misbahsy/doc-routerRust

jackalope

A Tauri desktop workspace for running coding agents side by side: parallel task queues in isolated Git worktrees, your existing agent sign-ins, and a review step before anything is merged.

Jev Tools & CLIs#community#jevbest#catalog-import

Community

Jackalope-Dev/jackalopeRust

triagedy

A Rust filter that reads JSONL security alerts on stdin and asks five Jev questions per alert — disposition, severity, false positive, escalation, ATT&CK class — then applies the routing policy in ordinary code you control.

Jev Tools & CLIs#community#jevbest#catalog-import

Community

m0rphtail/triagedyRust

XERJ

3.5k

An Elasticsearch-compatible autoindex search layer that optionally sends top hits to TypeSafe Jev and reorders results by relevance scores.

Jev Tools & CLIs#community#rust#search

Community

xerj-org/xerjRust

From the Jev guides

Original write-ups on what this entry is for.

Jev on the Gateways: Vercel, Workers AI, and OpenRouter

Oct 12, 2026

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

Exporting Decision Models: GGUF, ONNX, and Device Runtimes

Oct 10, 2026

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

Records at Rest: Documents, Inboxes, and Risk Queues

Sep 28, 2026

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

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai