How does Jev fit into an agent that already has a chat model?

It pairs a generating model with Jev as an optional companion router: Loki Autorouter classifies the first task of a session and picks a lower-cost model from the selected gateway, staying sticky for prompt caching.

What does the self-improving learning loop actually store?

The learning loop covers agent-curated memory with nudges, autonomous skill creation after complex tasks, FTS5 session search with summarization, and user modeling, compatible with the agentskills.io standard.

Can it run off my laptop without tying up my machine?

Seven terminal backends cover local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox, with installers for Linux, macOS, WSL2, Termux, and native Windows PowerShell.

What is the catch with switching models and platforms?

Use loki for chat, loki model to switch providers without code changes, loki gateway for messaging platforms, plus doctor, update, setup, tools, and config commands documented in the full docs.

On Linux, macOS, WSL2 or Termux run the curl install.sh one-liner; on native Windows run the PowerShell iex irm install.ps1 command. The installer handles uv, Python 3.11, Node.js, ripgrep, ffmpeg and a portable Git Bash under %LOCALAPPDATA%\loki on Windows.

How do I enable Jev alongside my chat model?

Choose Quick Setup + TypeSafe Jev in the CLI, or TypeSafe Jev (companion) during Loki Desktop onboarding, then paste your TYPESAFE_API_KEY. You can also run loki setup jev later, or use /jev status, /jev enable, /jev disable and /jev setup in an existing CLI session.

What is the Loki Autorouter?

An optional model router powered by Jev. It classifies the first task of a new session and selects a sufficiently capable lower-cost model from the currently selected gateway only, then keeps the route sticky for that session to preserve prompt caching.

CommunityJev Integrations & Adapters

Loki

Agent runtime with a terminal UI and a messaging gateway that keeps memory, self-built skills, session search, scheduling, and subagents across Telegram, Discord, Slack, WhatsApp, Signal, and CLI.

See highlights

Open github.com

Stars: 25
Language: Python
License: MIT
Published by: Community
Author: wundercorp
Last verified: Oct 2026

wundercorp/loki

Organization repository on GitHub

View on GitHub

The agent that evolves with you 𖤍

Forks: 6
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

The self-improving AI agent built by WunderCorp, Inc.. It's the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions. Run it on a $5 VPS, a GPU cluster, or serverless infrastructure that costs nearly nothing when idle. It's not tied to your laptop — talk to it from Telegram while it works on a cloud VM.

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

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Ranking Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

Works with OpenRouter, OpenAI, a custom endpoint, and other providers, switching with loki model and no code changes.
Closed learning loop with skill creation, skill self-improvement, memory nudges, session search, and user modeling.
One gateway process serves CLI plus messaging platforms with voice memo transcription and conversation continuity.
Scheduled automations, isolated subagents, RPC-scriptable tools, and batch trajectory tooling for research.

Watch out

Model and gateway keys are your own, first-run setup uses OpenRouter by default, and serverless or VPS hosting behavior depends on your infrastructure. 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

laya-jev-GraphRAG

GraphRAG layer that runs a 4-phase ingest-to-answer pipeline over Neo4j, Memgraph, AGE or Kuzu using Laya or Jev decisions instead of per-hop LLM calls.