LocalAI v5.0.0 Released: Major Overhaul of Chat and Agent Systems

At a Glance
| Item | Value |
|---|---|
| Repository | mudler/LocalAI |
| Version | v5.0.0 |
| Published | 2026-11-19 |
| License | MIT |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code when the information was collected. Dates are JST.
Overview
LocalAI is an open-source AI engine that allows you to run any model—including LLM, vision, audio, image, and video—on any hardware, even without a GPU. Its latest version, v5.0.0, has been released.
This update is massive, fundamentally revamping the chat functionality. The previous chat pages and agent execution environments have been replaced by a new server-side “Chats" system. This significantly changes the user experience, allowing conversations to persist even after closing browser tabs. If you are currently running a server, be sure to check the migration steps.
Breaking Changes and Deprecations
In v5.0.0, the chat and agent-related APIs have been significantly altered. Systems using existing APIs will stop working, so please refer to the migration table below.
| Old API / Option | New API / Option |
|---|---|
/api/agents |
/api/chats/agents |
/api/agents/skills |
/api/chats/skills |
/api/agents/git-repos |
(Removed) |
/api/conversations |
/api/chats |
local-ai agent command |
(Removed) |
LOCALAI_AGENT_POOL_* |
LOCALAI_COLLECTIONS_* |
LOCALAI_DISABLE_AGENTS |
LOCALAI_DISABLE_CHATS and LOCALAI_DISABLE_COLLECTIONS |
- API Changes: Endpoints such as
/api/agentsand/api/conversationswill return a404error. You must rewrite them to the new/api/chats/*endpoints. - Option Name Changes: Configuration item names have been updated. Old names will continue to work, but will trigger deprecation warning logs.
- Studio and Talk: These have been moved from standard features to “Add-ons." Administrators must install them as add-ons.
- Removal of liquid-audio: Configurations using
backend: liquid-audiowill no longer work. You must migrate to a configuration using LFM3-Audio-2B with thellama-cppbackend. - Simple Office Input Format: In the Office document creation tool, the traditional
blocksandsheetsarguments are now rejected. Please use Markdown or CSV moving forward.
These changes directly impact developers using LocalAI as an API server and users leveraging existing agent features.
Key Changes
Revamped Server-Side Chats
Conversation management is now handled on the server side. This allows chats to persist even if you close your browser tab. If a user sends a message while the model is working, it is queued and processed in the model’s next step. Additionally, models can now launch sub-agents in the foreground or background. There are no changes to the specifications of the existing /v1/chat/completions endpoint.
Introduction of Desktop App
A new Electron-based desktop app is provided for macOS (Apple silicon) and Linux (x86-64, ARM64). This replaces the previous launcher, allowing you to install, update, start, and stop the LocalAI server easily from a GUI. Beyond running the server on your own PC, you can also connect to and manage other LocalAI servers on your network.
Projects Feature
A “Projects" feature has been added, allowing you to group chats associated with specific folders. Each project can have its own files/ folder and a memory/ folder containing notes for the chat to reference. This enables consistent work based on specific contexts.
File Management and History
The new “Files" page allows for uploading, previewing, searching, and extracting archives. It also supports text search within PDF, Word, Excel, and PowerPoint files. Additionally, you can check changes made by files or models via the “History" tab and undo them if necessary.
Python Sandbox (Linux Only)
The “Sandbox" add-on allows you to execute Python or shell commands in an isolated environment (KVM or gVisor). This enables the model to perform tasks like creating graphs from CSV files or manipulating files. Network access is disabled by default to ensure security.
Expanded Add-on Library
An “Add-ons" library has been established to add new tools to chats, such as Slack or Discord bots, Google Drive, GitHub, and Home Assistant. Administrators can install necessary features from the library, and users can utilize those tools within their chats.
Releases Since Our Last Article
Compiled by Local Model Watch from the project’s GitHub releases: the versions between this release and the last one we covered, which did not get separate articles. Full history: release tracker.
| Version | Released | Release notes |
|---|---|---|
| v4.12.1 | 2026-10-23 | GitHub |
Related Articles
- LocalAI v4.12.0 Released: Web UI Refresh and Streaming Audio
- LocalAI v4.11.0 Released with Failover and NeMo Audio Models
- LocalAI v4.10.0 Released: Fleet Dashboard & M5 Support
What to Read Next
- Follow this tool → LocalAI overview and release history (139 releases tracked)
- Other inference engines and runtimes → llama.cpp / Ollama / vLLM
Sources
- mudler/LocalAI v5.0.0 Release Notes
- Chats: moving from the old chat, agents and skills
- Upgrading: Studio and Talk are add-ons now
- Move from the older launcher
- Simple Office creation tools take Markdown or CSV
- LFM3-Audio speech-to-speech on llama.cpp, remove liquid-audio
Supported Models and Hardware
In v5.0.0, speech-to-speech functionality using LiquidAI’s LFM3-Audio-2B is supported on the llama-cpp backend. This allows for audio input/output and simultaneous text and audio processing via the Realtime API (/v1/realtime, WebSocket, and WebRTC). When using realtime_audio with llama-cpp backend models, the setting pipeline.any_to_any: true is required. Additional features like the sandbox and desktop app are compatible with Linux (x86-64 and ARM64) and macOS (Apple silicon) environments.
How to Update
To perform a fresh installation of the server, run the following script:
curl -sSL https://localai.io/install.sh | sh
local-ai
If using a container environment like Docker, use the following command:
docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
If using the desktop app, download the binary for your platform from the GitHub releases page. When migrating from the old launcher, settings, models, and backends are preserved in the ~/.localai folder.

