About LibreChat 0.8.8-rc2 — Self-Hosted Multi-Model AI Chat
LibreChat is a self-hosted chat application for AI models. It looks and works much like ChatGPT, but you run it on your own server, connect it to whichever providers you choose, and give accounts to the people who should use it. It is MIT licensed and maintained by Danny Avila and contributors.
This page links to GitHub’s source archive of the v0.8.8-rc2 tag. Bineret did not build LibreChat and is not affiliated with it.
Know what you are downloading
Two things to be clear about before you start.
It is a release candidate. v0.8.8-rc2 came out on 3 September 2026. Every LibreChat release on GitHub is marked as a pre-release, including 0.8.7, so there is no “stable” tag to pick instead; rc2 is simply the newest. The project asks you to read its changelog for breaking changes before every update.
It is source code, not an installer. The zip is 15.3 MB and holds 4,830 files: the API, the React client, shared packages, Docker and Helm files. You run it with Docker or build it with Node.js. And one detail that catches people out: the docker-compose.yml in this archive pulls the prebuilt image librechat-dev:latest from LibreChat’s registry. It does not build the code you downloaded. To run exactly rc2, build from source.
What it does
- Many providers in one interface. Anthropic, OpenAI and Azure OpenAI, AWS Bedrock, Google and Vertex AI, plus any OpenAI-compatible API through custom endpoints. That covers local and remote options such as Ollama, OpenRouter, Mistral, DeepSeek and Groq.
- Agents. No-code agents with MCP servers, file search, code execution, reusable Skills and Subagents, shareable with chosen users and groups.
- Code Interpreter. Sandboxed execution in Python, Node.js, Go, C and C++, Java, PHP, Rust and Fortran, through a separate self-hostable service.
- Artifacts. React, HTML and Mermaid output rendered in the chat.
- Conversations you can work with. Presets, editing and resubmitting, branching and forking, search, and import from ChatGPT and Chatbot UI.
- Files, images and voice. Chat with uploaded files, generate and edit images, speech to text and text to speech.
- Multi-user. Accounts, authentication including OpenID and SAML, and a UI in dozens of languages, Persian among them.
What rc2 adds
Agent runs can be interrupted, steered and queued, and can pause for a human to approve a tool call. The Agent Builder is unified. Scheduled Chats arrive as an experimental feature. Projects are redesigned, and search covers full message text. Security headers are on by default, with an opt-in nonce-based CSP. New models include GPT-5.6, Claude Fable 5.1, Opus 5 and Sonnet 5, and Gemini 3.8, 3.7 and 3.6 Flash.
What it needs to run
The default docker-compose.yml starts six containers: LibreChat itself, an admin panel, MongoDB 8.0.20, Meilisearch 1.35.1 for search, pgvector on PostgreSQL 15, and a RAG API for retrieval over uploaded files. Without Docker you need Node.js, pinned to 24.16.0 in .nvmrc, and a MongoDB of your own. It listens on port 3080 by default.
You also need at least one model: an API key from a hosted provider, or a local model through something like Ollama.
When to use something else
If you are one person who wants a desktop chat app, running MongoDB and a search server is heavy for that. A desktop client such as LobeHub Desktop is less to look after. If you need a version that has been through a stable release process, note that LibreChat does not currently tag one.
Key capabilities
Every provider, one chat
Anthropic, OpenAI and Azure, Bedrock, Google and Vertex AI are built in. Anything that speaks the OpenAI API, from OpenRouter to a local Ollama, plugs in as a custom endpoint, and you can switch between them mid-chat.
Agents with real tools
Build agents without code and give them MCP servers, file search, code execution, reusable Skills and Subagents. In rc2 a run can be interrupted, steered, queued, or paused for a human to approve a tool call.
Built for more than one user
Accounts, OpenID and SAML sign-in, and agents and prompts shared with chosen users and groups. rc2 turns on default security headers and adds an opt-in nonce-based Content Security Policy.
A release candidate
v0.8.8-rc2 is the newest tag. LibreChat marks every release as a pre-release, 0.8.7 included, so there is no stable build to choose instead. Read the changelog before updating.
Docker pulls, it does not build
The docker-compose.yml in this tag runs the prebuilt librechat-dev:latest image, not the source in the folder. To run exactly rc2, build from source with npm ci, npm run frontend and npm run backend.
What the stack costs you
Six containers by default: the app, an admin panel, MongoDB, Meilisearch, pgvector and a RAG API. That is a small server's worth of services, fine for a team and heavy for one person.
Questions & Answers
Is this a stable release?
No, it is release candidate 2 of 0.8.8. LibreChat marks every GitHub release as a pre-release, 0.8.7 included, so there is no stable tag to choose instead. Read the changelog before upgrading an existing install.
Why does Docker not run the code I downloaded?
The docker-compose.yml in this tag uses the prebuilt image registry.librechat.ai/danny-avila/librechat-dev:latest. It does not build from the folder. To run exactly rc2, build from source with npm, or build your own image from the included Dockerfile.
Do I need all six containers?
MongoDB is required. Meilisearch powers conversation search, and pgvector with the RAG API power retrieval over uploaded files; the admin panel is a separate management interface. The compose file starts them all by default.
Can it use local models?
Yes. Any OpenAI-compatible API works as a custom endpoint, which includes local servers such as Ollama.
Can several people use one install?
Yes. It is multi-user, with registration, OpenID and SAML sign-in, and agents and prompts that can be shared with chosen users and groups.
Is it free for commercial use?
The MIT licence allows commercial use, modification and redistribution. The model providers you connect bill you separately for what you use.
Is Bineret connected to LibreChat?
No. This page describes LibreChat and links to the source archive GitHub generates for the project's tag.
Tutorials
Before you start
You need Docker with Compose, or Node.js 24.16.0 (the version in .nvmrc) plus a MongoDB server. You also need at least one model: an API key from a provider, or a local model server such as Ollama.
Unpack and configure
unzip LibreChat-0.8.8-rc2.zip
cd LibreChat-0.8.8-rc2
cp .env.example .envOpen .env and fill in CREDS_KEY, CREDS_IV, JWT_SECRET and JWT_REFRESH_SECRET, which ship empty. They protect stored credentials and sign logins, so use long random values and keep them out of version control. Add the API keys for the providers you want. LibreChat's documentation has a generator for the credential values.
Run with Docker
docker compose up -dThen open http://localhost:3080. This starts LibreChat, the admin panel, MongoDB, Meilisearch, pgvector and the RAG API. Note that the compose file pulls the prebuilt librechat-dev:latest image rather than building the source in this folder, so what runs is the registry's latest build, not necessarily rc2.
Run exactly this version, from source
npm ci
npm run frontend
npm run backendMONGO_URI in .env must point at a running MongoDB; the example value is mongodb://127.0.0.1:27017/LibreChat.
Create the first account
Registration is open by default (ALLOW_REGISTRATION=true). Register yourself, then decide whether to set it to false so strangers who find the address cannot sign up. Accounts can also be created from the shell:
npm run create-userConfigure providers and features
Endpoints, model lists, agents and interface options live in librechat.yaml. Start from librechat.example.yaml in the archive.
Support
Where support comes from
LibreChat is maintained by Danny Avila and contributors. Start with the documentation and the changelog. Report bugs as GitHub issues; the project also runs a Discord community. For new features it asks you to open an issue and discuss before sending a pull request.
What Bineret covers
This page. If the download link is broken or something written here is wrong, tell us and we will fix it. We do not host, install or configure LibreChat, and we do not supply model API keys.
No warranty
The MIT licence provides the software as-is, and this is a release candidate. Back up MongoDB before upgrading, and keep the four credential secrets from .env somewhere safe; stored credentials cannot be decrypted without them.