AI Chat Clients

Open WebUI156.5k★

Open WebUI is a ChatGPT-style web app you run on your own server. Point it at Ollama, LM Studio, vLLM or any OpenAI-compatible API, and your team gets chat, document Q&A (RAG: the AI answers from your files), web search, voice calls and per-group permissions. It's free to run. Setup is pip install open-webui or one docker run. The models behind it are the real hardware or API bill. LDAP, SSO and SCIM ship free. Releases land every week or two, so pin versions. The paywall is mostly about the logo. Keep the Open WebUI branding and use is free at any size. Past 50 end users in a rolling 30 days, white-labeling needs an enterprise license, and so does Terminals, the per-user sandbox add-on. Prices aren't published. Solo and small teams: free. Larger teams: free unless you rebrand or want Terminals and an SLA. The catch: security. The project has published more than 150 advisories since February, many rated high: account takeover, session-token theft, SSRF. Most need a logged-in user, so patch fast and keep it off the open internet. Also, the license isn't OSI open source; LibreChat is the MIT-licensed comparison.

lobehub83.4k★

LobeHub is what LobeChat grew into: one chat app for OpenAI, Claude, Gemini and local models that now runs teams of AI agents on a schedule, with memory and a marketplace of MCP plugins. It's a self-hosted answer to ChatGPT, in the same lane as Open WebUI and LibreChat, and free to run. Self-hosting got heavier with 2.0. Browser-only mode is gone, so even the one-click Vercel deploy needs Postgres, S3-style storage and auth secrets. You also pay your own model API bill. The hosted cloud meters credits. Free buys roughly 40 to 110 messages a month on cheap models and none on flagships. Starter is $12.90/mo ($9.90 billed yearly), Premium $24.90, Ultimate $49.90. Solo: self-host, or Starter to skip the server. Small teams: self-host. Larger teams: Enterprise. The catch is security. CVE-2026-23733 (critical, 9.6) let a chat message rendered as a Mermaid diagram run attacker code, reaching command execution in the desktop app; fixed in 2.0.0-next.180. A similar bug followed (fixed in 2.1.48), and an October 17 advisory showed any logged-in user could execute code on the server (fixed in 2.1.49). And the license is Apache 2.0 plus conditions: shipping a modified version needs a paid commercial license.

cherry-studio52.9k★

Cherry Studio puts every AI model behind one desktop window: chat, agents, knowledge bases, image generation and translation on Windows, macOS and Linux. Bring your own keys for OpenAI, Anthropic or Gemini, or run local models through Ollama and LM Studio. The app is free under AGPL-3.0. There's no server to run. Download, paste an API key, go. Chats, keys and knowledge bases stay on your machine, and requests go straight to the provider you picked. iOS and Android apps exist too. Two things cost money. Cherry Flash is an optional $8/month plan (a promo off $10) that bundles DeepSeek, Qwen and GLM usage so you skip the keys. Enterprise is a hosted team workspace: prepay US$10,000 in tokens and you get a one-year license with unlimited seats, paid by Alipay or bank transfer. Solo and small teams: free plus API costs. Larger teams: that Enterprise math, or Open WebUI or LibreChat. The catch: security history. Cherry Studio has shipped several remote-code-execution bugs, from one-click malicious links to malicious MCP servers to a search flaw that ran attacker code with full system access, fixed only in v2. Stay current. Anonymous usage telemetry is on until you turn it off.

LibreChat46.2k★

LibreChat is a self-hosted ChatGPT replacement: one chat window for Claude, GPT, Google, Bedrock, Ollama and any OpenAI-compatible API, with agents, MCP tools, artifacts and search built in. It's MIT licensed with no paid tier, and ClickHouse bought it in November 2025, so a company now funds the work. Setup is Docker Compose, but it's a real stack: the app, an admin panel, MongoDB, Meilisearch, Postgres with pgvector and a RAG API. Running code needs a separate service you deploy yourself. Releases ship constantly, and the README says to check the changelog for breaking changes before upgrading. You bring your own API keys, so the bill is model spend plus a server. Login covers OAuth, SAML, LDAP and two-factor. Solo: free, worth it if you want every model behind one login. Small teams: free, one deployment with shared agents. Larger teams: free, but budget engineer time for patching. The catch is security. LibreChat has published nearly 30 advisories since mid-2025, three rated critical in 2026, including one where any registered user could run shell commands as root inside the container. All are patched. This server holds everyone's API keys, so stay current and keep it off the open internet.

onyx32.9k★

Onyx is a self-hosted ChatGPT that knows your company. It indexes Slack, Google Drive, Confluence, Jira, GitHub, SharePoint and dozens more, then answers questions with citations back to the source, plus agents, deep research and web search. It works with any LLM, local ones included, and the Community Edition is MIT. Self-hosting has two modes. Lite is a chat UI that runs in under 1 GB of memory but cannot index documents, and indexing is the whole point. Standard is the real thing: OpenSearch, background workers, model servers, Redis and Postgres, with a minimum of 4 vCPU and 10 GB of RAM. Install is one script; keeping the index healthy is ongoing work. Business is $20/user/month billed annually or $25 monthly, on Onyx Cloud or self-hosted. It adds RBAC, permission inheritance, query history and usage dashboards. SAML SSO, SCIM and white-labeling are Enterprise, priced by quote. Solo and small teams sharing the same docs: free. Anyone with sensitive docs: Business. The catch: this app holds every credential you connect, and 2026 brought a run of authorization bugs, including a critical one that leaked other users' OAuth tokens. Patch promptly. Open WebUI is lighter if you only need chat.

5ire5.4k★

5ire is a desktop chat app for Windows, macOS and Linux that puts OpenAI, Anthropic, Gemini, DeepSeek, Ollama and more behind one window, with MCP tools and a local document knowledge base built in. Bring your own API keys. The app is free. Install is a normal download, and chat works right away. MCP tools need Python, Node.js and uv installed first, which is where non-developers stall. The knowledge base indexes PDF, Word, Excel, PowerPoint, text and CSV files on your machine with the bge-m3 embedding model. Nothing is paywalled for personal or in-house use. The license is Apache 2.0 plus extra terms: ship a modified build to enterprise clients over 10 users or inside hardware you sell, or use the 5ire name, and you need a commercial license. Solo and small teams: free plus API costs. Larger teams: no admin console, so look at Open WebUI or LibreChat. The catch: five critical remote-code-execution bugs in eight months, each rated 9.6. Each turned something rendered in the chat window (a message, a Mermaid diagram, a chart) into code running on your machine. The two January bugs were fixed in 0.15.3. Nothing has shipped since March 2026, including fixes already merged.

talorys1.4k★

Talorys is a personal AI assistant that lives in your own Cloudflare account instead of someone else's servers. You get chat with long-term memory, tasks, notes, projects, and reminders that fire on schedule as push notifications to your phone or laptop. It's MIT-licensed, single-user, and sends nothing to its developer. Setup is one command, npx create-talorys@latest. It logs into Cloudflare through Wrangler's OAuth, asks for an owner password, deploys a private Worker plus a Pages site, and prints your URL. No server to patch, no database to run: everything lives in one SQLite-backed Durable Object, and the Worker has no public URL. It's built for Cloudflare's free plan. The chat model is GLM-4.7-flash on Workers AI, and the free allowance is 10,000 Neurons a day, roughly 1.8 million input or 275,000 output tokens on that model. Past that, chat pauses until the daily reset unless you're on Workers Paid ($5/month minimum) at $0.011 per 1,000 Neurons. Solo: free. Teams: skip it, it's single-user by design. The catch: you're stuck with Workers AI's open models, no Claude or GPT without writing a provider yourself, and Cloudflare processes every message. Compared with OpenClaw or ChatGPT's memory, this trades capability for privacy.