jan-cargo-vendor-v0.8.8.tar.gz

jan-dependency-tarballs-v0.8.8.sha256

jan-mac-universal-0.8.8.zip

jan-yarn-cache-v0.8.8.tar.gz

Jan.app.tar.gz

Jan_0.8.8_amd64.AppImage

Jan_0.8.8_amd64.deb

Jan_0.8.8_arm64-setup.exe

Jan_0.8.8_universal.dmg

Jan_0.8.8_x64-setup.exe

latest.json

## Jan v0.8.8: security fixes, llama.cpp 0.7.0, and Trusted Hosts for the local API

A security update to v0.8.7. We recommend updating. Your data and settings carry over; a few behaviours change, listed under **What you might notice**.

### Security

- **Links in chat open in your browser.** Links in model replies, tool results, model cards and release notes now always open in your system browser, never inside Jan's window. Links that aren't web or email links are shown as plain text.
- **The local API server checks where browser requests come from.** A web page can only call it if its origin is trusted: the same machine, Jan itself, or a host you list in **Trusted Hosts**. Requests without a browser origin (SDKs, scripts, curl, Claude Code) work as before.
- **Server-side tool execution stays opt-in for every endpoint.** Requests to the local API run MCP tools only when **Settings > Local API Server** allows server-side tool execution.
- **Paths are checked before Jan reads or writes them.** Thread and assistant ids, and the file operations extensions use, stay inside Jan's data folder.
- **The Claude Code integration writes your shell profile safely.** Values are quoted, unusual variable names are refused, and Jan's lines are kept in one marked block that Reset removes completely without touching your own lines.
- **The agent's file tools no longer follow a link out of their workspace**, and folders such as your home, credentials and startup folders can't be attached as writable.
- **Windows: Jan runs system tools from the Windows system folder**, and sign-in links are opened without going through `cmd`.
- **MCP sign-in only uses web addresses** (`http`/`https`) for its endpoints.

### What you might notice

- **Trusted Hosts for browser-based clients.** A web app on another machine (for example Open WebUI's direct connection) or a browser extension calling Jan's API needs its host added to **Settings > Local API Server > Trusted Hosts**. Entries can include a scheme, so `app://obsidian.md` works for Obsidian plugins (thanks to [@SnowingFox](https://github.com/SnowingFox), [#8608](https://github.com/janhq/jan/issues/8608)). Pages without a normal origin (local files, sandboxed frames) can be allowed by listing `file://` or `null`.
- **The CORS switch now takes effect**, and it starts **on** after updating. In earlier versions it was shown but had no effect.
- **Custom environment variables for Claude Code:** keys that change how programs are found (`PATH`, `PATHEXT`, `SystemRoot`, `ComSpec`, `LD_*`, `DYLD_*`, `JAN_*` and the GPU runtime paths Jan uses) are refused with a message. Other variables, including `NODE_OPTIONS`, save as before.
- **Moving the data folder** to a system folder, your home folder itself or a drive root (for example `D:\`) is refused; choose a folder such as `D:\Jan`.
- **MCP servers whose sign-in endpoints aren't web addresses** now show a sign-in error.

### New

**llama.cpp engine 0.7.0.**

The bundled engine moves from 0.5.0 to 0.7.0 (b11669). New per-model settings: an **MoE expert GPU cache** for mixture-of-experts models, **Probabilistic draft sampling** and **Draft batch size** for speculative decoding. They default to the engine's own defaults, so nothing changes unless you set them. Saved prompt caches from the previous engine are discarded once and rebuilt. ([#9399](https://github.com/janhq/jan/pull/9399))

### Fixes

**Local models failed to load on Windows when the user name had non-English characters** (for example Cyrillic). ([#9418](https://github.com/janhq/jan/pull/9418))

**With an HTTP proxy configured, documents with "Use embeddings" failed**, because Jan's requests to its own engine went through the proxy. They now go direct; remote providers still use the proxy. ([#9353](https://github.com/janhq/jan/pull/9353))

**An attachment was sent again on every turn**, adding a copy per message and defeating the prompt cache. It's now included once per request. Thanks to [@yaodong-shen](https://github.com/yaodong-shen). ([#9358](https://github.com/janhq/jan/pull/9358))

**With Shell turned off, a model that called the shell ended its reply.** It now gets "Shell is turned off" and the reply continues. ([#9319](https://github.com/janhq/jan/issues/9319))

**macOS: quitting Jan left the MLX server running** with its model in memory. It's stopped on quit, and one left by an earlier Jan is stopped at startup; an MLX server you started yourself is left alone. ([#9355](https://github.com/janhq/jan/pull/9355))

### Thanks

Fixes in this release include work from [@greatjourney589](https://github.com/greatjourney589) ([#8068](https://github.com/janhq/jan/pull/8068)), [@SnowingFox](https://github.com/SnowingFox) ([#8627](https://github.com/janhq/jan/pull/8627)) and [@orenyomtov](https://github.com/orenyomtov).

### Known issues

- **Auto compact** needs **Max Context Tokens** set for remote models; Jan does not detect a remote model's context window yet ([#9317](https://github.com/janhq/jan/issues/9317)).
- Local models do not run on Intel Macs, as in v0.8.5.

Full notes: https://github.com/janhq/jan/blob/main/docs/src/pages/changelog/2026-10-28-jan-v0.8.8.mdx

**Full Changelog**: https://github.com/janhq/jan/compare/v0.8.7...v0.8.8


0.8.8

Jan v0.8.8 is a security release: links from chat open in your browser, the local API checks where requests come from, and several file, path and shell-profile checks are tightened. It also updates the bundled llama.cpp engine to 0.7.0 and fixes local models on Windows user names with non-English characters.

Jan v0.8.8: Security Fixes, llama.cpp 0.7.0 & Local API Trusted Hosts

Jan v0.8.7 adds a switch to turn the shell tool off in chat and a separate image size limit, keeps project threads in their project when you attach a document, leaves pasted code with @decorators as typed, removes the hosted Exa MCP default, and brings built-in subagents to the Jan agent.

Jan v0.8.7: Shell Switch, Image Size Limit, Project Threads & Agent Subagents

Jan v0.8.6 is a hotfix for v0.8.5: the Linux AppImage no longer crashes on model load, NVIDIA Turing GPUs (GTX 16xx, RTX 20xx) run the CUDA engine again on older drivers, Qwen3.8 with MTP is fast again when Parallel Sequences is auto, and images returned by MCP tools reach the model as images.

Jan v0.8.6: AppImage, Turing GPU, MTP Speed & MCP Image Fixes

Jan v0.8.5 is a fixes release: the llama.cpp engine (now 0.5.0) ships inside the app instead of being downloaded, the jan CLI is no longer bundled, the desktop executable is renamed to Jan-Desktop, the Local API Server enforces Trusted Hosts on 0.0.0.0, and a batch of llama.cpp, MCP, platform and agent fixes. No new features are announced in it.

Jan v0.8.5: llama.cpp Stability, MCP Hardening & Platform Fixes

Jan v0.8.4 adds native web_search / web_fetch tools, per-model chat-template kwargs for llama.cpp, a token counter for remote and MLX providers, and moves settings and credentials into a backend-managed store (localStorage kept as a downgrade snapshot) - alongside a batch of chat, MCP, provider, and streaming fixes.

Jan v0.8.4: Native Web Search, Per-Model Chat Templates & a Backend Settings Store

Jan v0.8.3 brings navigable message version branching, a fully custom Linux titlebar that follows your DE's button layout, a unified reasoning/tool chain-of-thought timeline, interactive HTML and static SVG artifact previews, video input for local vision models, and a sweeping batch of streaming performance, citation, sampler, and error-surfacing fixes.

Jan v0.8.3: Message Branching, Linux Custom Titlebar & HTML/SVG Artifacts

Jan v0.8.2 adds AMD ROCm/HIP backend support on Linux, a branded phased-startup loader with work deferred past first paint, pause/resume model downloads, a safer default context size, and a batch of chat, provider, and toast fixes.

Jan v0.8.2: Faster Startup, AMD ROCm/HIP on Linux & Resumable Downloads

Jan v0.8.1 adds Anthropic-compatible custom providers, OS-native TLS trust, a composer-anchored sampler popover with capability gating, persistent per-thread and per-message errors, and a major llama.cpp settings overhaul with split version/backend selectors, file-backed persistence, and per-model context sizing.

Jan v0.8.1: Anthropic-Compatible Custom Providers, Per-Message Errors & llama.cpp Settings Overhaul

Jan v0.8.0 ships Multi-Token Prediction for compatible llama.cpp models, a unified llama.cpp router process, inline MCP tool approval with citation cards, per-model fit labels, bulk model deletion, per-thread activity indicators, and a backend dependency checker.

Jan v0.8.0: Multi-Token Prediction, llama.cpp Router Mode & Inline MCP Approval

Jan v0.7.9 fetches the latest models during onboarding, caps context length to avoid high RAM usage, fixes CLI on Windows, and improves data location management.

Jan v0.7.9: CLI on Windows, Smarter Context Management & Safer Data Location Management

Jan v0.7.8 brings OpenClaw integration, a brand-new Jan CLI, and automatic context management so you never get cut off mid-conversation.

Jan v0.7.8: OpenClaw Integration, Jan CLI & Auto Context Management

Jan v0.7.7 delivers native MLX support for macOS, app-wide UI refresh, file uploads in Projects, and improved API server capabilities.

Jan v0.7.7: Native MLX Support & UI Refresh

Jan v0.7.6 delivers a completely reworked chat interface, plus introduces the Jan V3 model in onboarding.

Jan v0.7.6: Unified Interface & Jan V3 Model

Jan v0.7.5 is a hotfix release that addresses model import issues on Windows, vision model compatibility, CUDA backend installation, and MCP server cleanup.

Jan v0.7.5: Hotfix Release

Jan v0.7.4 delivers a simplified onboarding experience, powerful file attachment improvements, and smarter browser integration for seamless AI interactions.

Jan v0.7.4: Streamlined Onboarding & Enhanced File Attachments

Jan v0.7.3 introduces powerful browser automation with Jan Browser MCP, proactive AI assistance, and refined UX improvements for a smoother experience.

Jan v0.7.3: Enhanced Browser Control & Smarter Interactions

Jan v0.7.2 updates the happy-dom dependency to v20.0.0 to address a recently disclosed sandbox vulnerability.

Jan v0.7.2: Security Update

Jan v0.7.1 focuses on bug fixes, including a windows version revert and improvements to OpenRouter models.

Jan v0.7.1: Fixes Windows Version Revert & OpenRouter Models

Jan v0.7.0 introduces Projects, model renaming, llama.cpp auto-tuning, model stats, and Azure support.

Jan v0.7.0: Jan Projects

Use AI in teaching without risking student data. Jan helps teachers plan lessons, grade faster, and communicate with parents.

AI for teachers who care about student privacy

It's possible to use AI without risking client data. Jan helps lawyers save time while keeping clients safe.

Private AI for legal professionals who need confidentiality

Check if ChatGPT down right now, and learn how to use AI that never goes down.

If ChatGPT is down, switch to AI that never goes down

See the best ChatGPT alternatives in 2025. We've listed tools that are alternatives to ChatGPT.

ChatGPT alternatives that actually replace it

New experimental Auto Optimize feature, custom llama.cpp backend support, vision model imports, and critical bug fixes

Jan v0.6.10: Auto Optimize, custom backends, and vision model imports

Major multimodal support with image uploads, MCP out of experimental, auto-detect model capabilities, and enhanced tool calling

Jan v0.6.9: Image support, stable MCP, and powerful model tools

Llama.cpp stability upgrades, Linear/Todoist MCP tutorials, new model pages (Lucy, Jan‑v1), and docs reorganization

Jan v0.6.8: Engine fixes, new MCP tutorials, and cleaner docs

Full support for OpenAI's open-weight gpt-oss models and new Jupyter MCP integration guide

Jan v0.6.7: OpenAI gpt-oss support and enhanced MCP tutorials

Complete 5-minute beginner guide to running OpenAI's gpt-oss locally. Step-by-step setup with Jan AI for private, offline AI conversations.

Run OpenAI's gpt-oss locally in 5 mins (Beginner Guide)

A simple guide to replicating Deep Research results for free, with Jan.

Replicating Deep Research in Jan

Major llama.cpp improvements, Hugging Face provider support, and refined MCP experience

Jan v0.6.6: Enhanced llama.cpp integration and smarter model management

New MCP examples, updated pages, and bug fixes!

Jan v0.6.5 brings responsive UI and MCP examples!

Unlocking MCP for everyone and bringing our latest model to Jan!

Jan v0.6.3 brings new features and models!

Jan v0.6.1 is here: It's a whole new vibe!

Jan v0.5.17 is out: Qwen3 support is now more reliable

Qwen3 support is now more reliable.

If you're running Qwen3-30B-A3B locally, don't guess your way through the settings. This guide tells you what actually works based on Qwen's own documentation and what we've seen hold up in practice.

Best Settings to Run Qwen3-30B-A3B Locally

Jan v0.5.16 is out: Security fixes and major improvements to Model Hub and chat experience

Security fixes and UI improvements

Jan v0.5.15 is out: Advanced llama.cpp settings and cloud model support

You can now tweak llama.cpp settings, and add any cloud model!

ChatGPT requires internet to work. Jan is a free offline ChatGPT replacement that runs 100% on your computer. Download for Mac, Windows & Linux.

Offline ChatGPT: You can't run ChatGPT offline, do this instead

A straightforward guide to running DeepSeek R1 locally regardless of your background.

Run DeepSeek R1 locally on your device (Beginner-Friendly Guide)

A straightforward guide to running AI models locally on your computer, regardless of your background.

How to run AI models locally as a beginner?

Jan v0.5.14 is out: Run DeepSeek R1 Distills error-free!

Run DeepSeek R1 Distills error-free!

Jan v0.5.13 is here: A few key issues have been solved.

A few key issues have been solved!

Improved Privacy settings to give full control over analytics

Jan gives you full control over your privacy

Critical issues fixed, Mac installation updated.

Jan v0.5.11 is here!

Jan is faster, smoother, and more reliable.

Jan v0.5.10 is live

Jan v0.5.9 is here: fixing what needed fixing.

Model downloads & running issues fixed

Jan v0.5.8 is out: Jan supports Qwen2.5-Coder 14B & 32B through Cortex

Jan supports Qwen2.5-Coder 14B & 32B

This release is mostly focused on bug fixes.

Jan has Stable, Beta and Nightly versions

Jan is more stable 👋

Jan is faster now

v0.5.3 is out with stability improvements!

Jan supports Claude 3.5 Sonnet

Jan has integrated NVIDIA NIM and supports Qwen 2 7B

Jan supports NVIDIA NIM

Revamped Jan's UI to make it clearer and more user-friendly

New UI & Codestral Support

Jan now supports Cohere's Aya 23 8B & 35B and Microsoft's Phi-3-Medium.

Jan now compatible with Aya 23 8B & 35B and Phi-3-Medium

Jan now supports more GGUF models

Jan has added compatibility with Llama3 & Command R+

Jan now supports Llama3 and Command R+

Jan has a new Mistral Extension letting you chat with larger Mistral models via Mistral API

New Mistral Extension

Jan connects to Groq's API, with VRAM-aware model recommendations, a better hardware troubleshooting guide and fixes to the data folder.

Groq API Integration

The Nitro TensorRT-LLM extension arrives for faster inference on NVIDIA GPUs, with an updated models.json.

Nitro-Tensorrt-LLM Extension

Import models straight from the app, a revamped system monitor, Homebrew installs on Apple Silicon, and a fix for HTTP proxy download errors.

Import models directly using the UI

Run Jan on your own home server with Helm, headless or with a web UI, plus Intel Arc and AMD GPU support through Vulkan and 48 fixes.

Run Jan on your home-servers with Helm

Move Jan's data folder anywhere, even an external drive, reset to factory settings, and see when each thread was last updated.

Jan Data Folder

A dashboard for Jan's local API server, HTTP proxy support for downloads, a reorganized settings page and smoother auto-updates.

Local API server

More thread settings in the right panel, switching models mid-thread, a check for whether a model runs on your device, and stability fixes.

Thread settings options in the right panel

Faster inference on Mac, Windows, Linux and GPUs, new Trinity and Pandora models, and remote OpenAI models like GPT-4 with your API key.