What it is
LibreChat is a self-hosted chat app that looks and works much like ChatGPT but puts models from OpenAI, Anthropic, Google, AWS Bedrock, Azure, local servers and others behind one interface and one sign-in.
ClickHouse acquired the project in November 2025, said it stays MIT-licensed, and the repository moved from danny-avila to LibreChat-AI. Founder Danny Avila joined ClickHouse. The paid hosted Code Interpreter API was wound down after the deal; code execution now runs on ClickHouse/code-interpreter, an Apache 2.0 sandbox you host yourself.
Version 0.8.8, released on 30 September 2026, added the ability to interrupt and steer a running agent, approve tool calls, a beta Agents API, experimental agent plugins and scheduled chats (beta, off by default).
Our review
LibreChat is the self-hosted chat front end with the fewest licence strings attached, and that matters more than any single feature. It is MIT, so a company can fork it, rebrand it and ship it internally without asking anyone. Shopify, quoted in ClickHouse's acquisition post, runs an internal fork with more than 30 MCP servers connected and merges most changes back upstream; Daimler Truck offers it to all employees.
The agent work has moved quickly. In our reading of the changelog, v0.8.6 brought skills and sub-agents and v0.8.8 added mid-run steering, tool-call approval and a beta Agents API that speaks Chat Completions and Open Responses. Model support tracks the frontier closely: the September release already listed GPT-6, Claude Opus 5.5 and Gemini 3.8 Flash.
The cost is operational. The default Docker Compose file starts MongoDB, Meilisearch for conversation search and a RAG API backed by PostgreSQL with pgvector, and most behaviour lives in librechat.yaml plus a long .env file. Registration is open by default and the first account becomes admin, so an exposed instance needs ALLOW_REGISTRATION turned off or SSO in front before anyone shares the link.
Security is the other line item. The repository published 24 security advisories in 2026 alone, including a high-severity SSRF through user-supplied endpoint URLs (fixed in v0.8.4-rc1) and a flaw in the MCP OAuth flow. That volume partly reflects an active, disclosed process, but it means pinning versions and patching monthly is part of running it. Open WebUI has the friendlier admin screens; LibreChat has the cleaner licence and the more ambitious agent roadmap.
Where it earns its place
An IT lead at a 500-person firm already paying for Azure OpenAI and Anthropic
Give staff one internal ChatGPT-style app behind company SSO instead of separate vendor seats
Run the Docker Compose stack or the Helm chart, wire OpenID Connect to Entra ID and list the permitted models in librechat.yaml. Spend becomes pay-per-token API usage rather than per-seat subscriptions; the trade is one engineer owning upgrades and patches.
A data platform team on ClickHouse
Let analysts ask questions of the warehouse in plain English
Connect the ClickHouse MCP server to a LibreChat agent. ClickHouse says its own internal deployment handles about 70% of data warehouse queries for more than 200 users this way.
A product team shipping an internal agent to non-technical colleagues
Package tools, skills and approval steps without building a front end
Build the agent in the Agent Builder, attach MCP servers, and turn on tool-call approval so a person confirms writes. The beta Agents API exposes the same agent to other apps, though beta endpoints may change between releases.
Before you commit
- Setup
- Clone the repo, copy .env.example and run docker compose up -d; the docs say the first pull takes a few minutes and the app answers on port 3080. Minimum is 1 vCPU and 1 GiB RAM, with at least 2 GB given to Docker for all services; no GPU unless you also run local models.
- Integrations
- OpenAI, Anthropic, Google, AWS Bedrock, Azure OpenAI and any OpenAI-compatible endpoint; MCP servers, file search via the bundled RAG API, self-hosted code execution, Langfuse tracing and a Helm chart for Kubernetes.
- Data and privacy
- Conversations, files and users sit in your MongoDB and Postgres; prompts go only to the model providers you configure. OpenTelemetry tracing is off by default and we found no vendor analytics in the default configuration. 24 advisories were published in 2026, so track releases.
- Cost at scale
- Software is free. Real cost is model API usage plus a small VM or Kubernetes namespace for the app, MongoDB, Meilisearch and Postgres; code execution needs a separate self-hosted sandbox service since the paid hosted API was discontinued.
- Limits
- About 1,000 open issues. Upgrades can change YAML and env settings (v0.8.8 points to three config changelogs), Apple Silicon needs an older MongoDB image, and scheduled chats and code workspaces are labelled experimental.
What's good
- MIT licence: fork and rebrand freely
- OpenAI, Anthropic, Google, Bedrock, Azure and local models in one UI
- Agents with MCP, skills, sub-agents and tool approval
- OAuth, OpenID Connect and LDAP sign-in included
What's not
- Needs MongoDB, Meilisearch and Postgres/pgvector for the full feature set
- Configuration split across librechat.yaml and a long .env
- Open registration by default; first user becomes admin
- 24 security advisories in 2026, so frequent patching
Best for
Companies that want a rebrandable internal ChatGPT alternative across many providers.
Alternatives
LibreChat: common questions
- Is LibreChat worth it?
- A permissively licensed company chat portal with serious agent and MCP features, provided someone on staff is happy to own a YAML file and a MongoDB. We rate it 4.4 out of 5.
- How much does LibreChat cost?
- Free and open source (MIT). No paid tier; you pay your model providers and hosting.
- What are the downsides of LibreChat?
- Needs MongoDB, Meilisearch and Postgres/pgvector for the full feature set; configuration split across librechat.yaml and a long .env; open registration by default; first user becomes admin; 24 security advisories in 2026, so frequent patching.
- Who is LibreChat best for?
- Companies that want a rebrandable internal ChatGPT alternative across many providers.
- How much does LibreChat cost for a team or in production?
- Software is free. Real cost is model API usage plus a small VM or Kubernetes namespace for the app, MongoDB, Meilisearch and Postgres; code execution needs a separate self-hosted sandbox service since the paid hosted API was discontinued.
- Is LibreChat safe for company data?
- Conversations, files and users sit in your MongoDB and Postgres; prompts go only to the model providers you configure. OpenTelemetry tracing is off by default and we found no vendor analytics in the default configuration. 24 advisories were published in 2026, so track releases.
- What are the best LibreChat alternatives?
- The closest alternatives we have reviewed are Open WebUI, LobeHub (formerly LobeChat) and AnythingLLM.
Answers checked Oct 23, 2026.
Make LibreChat? Show this rating on your site
