Comparison · Updated October 2026
Open WebUI vs LibreChat — and the one you don’t have to host
All three let you bring your own API keys to GPT, Claude, Gemini and more. Open WebUI and LibreChat are free apps you install and run yourself, and very good ones. PolyCog is a website you open, where the models argue their way to one answer.
- Open WebUI: Install Python 3.11, pip install open-webui, open-webui serve, Create the admin account, Add an API key, then Ask: first answer (6 steps)
- LibreChat: Install Git and Docker, git clone the repo, Copy .env.example to .env, docker compose up -d, Open localhost:3080, register, Add an API key, then Ask: first answer (7 steps)
- PolyCog: Open polycog.io, then Ask: first answer (2 steps)
Open WebUI from docs.openwebui.com (its pip route, the shortest it documents) · LibreChat from librechat.ai/docs (Docker Compose, its recommended route) · read October 2026. PolyCog's first answer runs on our shared keys (sample mode, three a day); your own keys take a sign-in and a paste.
Setup, step by step. Open WebUI: install Python 3.11, pip install open-webui, run open-webui serve and open localhost:8080, create the admin account, add an API key, ask (six steps). LibreChat: install Git and Docker, clone the repo, copy .env.example to .env, docker compose up -d (MongoDB, Meilisearch, a RAG API, a vector database), open localhost:3080 and register, add an API key, ask (seven steps). PolyCog: open polycog.io, ask (two steps). From a phone, both self-hosted apps also need a machine that stays on and an address with HTTPS, and their updates and backups are yours.
The short version
Open WebUI and LibreChat are the two big free chat apps you host yourself, and they are very good at it: any model with an API, local models through Ollama, agents, tools and team sign-in, with your chats on hardware you control. The price is running them: install, give them an address and HTTPS to reach them from your phone, and keep them updated. Both can show several models side by side, and Open WebUI can merge the answers; in neither do the models read and answer each other. PolyCog is a website: nothing to install, the same API keys at the same list prices, and the models debate, have their claims checked against web sources, and a flagship writes one synthesis. If you want your own server and everything on it, pick one of them. If you want several models to argue a question out on any device, that’s us.
Who pays for what
The same keys at the same list prices in all three. The difference is what you run to use them.
Open WebUI & LibreChat
Free software · your keys · your server
You→ Your server→ AI providers
The software costs nothing and the keys are yours, so the AI is billed at list price, exactly as in PolyCog. What you add is the machine it runs on and the hours to keep it running.
PolyCog
BYOK · free tier, $10/mo or $250 once, for the interface only
You→ AI providers, directly
The same keys at the same list prices, with nothing to run. A debate makes more calls than a chat, and the estimate before each run says how many cents.
What it takes to run them
From each project's own docs on the date above. Both are free to download; this is the rest of the bill.
| Open WebUI | LibreChat | |
|---|---|---|
| Install | pip (Python 3.11), Docker, Kubernetes or a desktop app | Docker Compose (Git + Docker), or Node.js with your own MongoDB |
| What runs | One app; the slim image idles at about 300 MiB of RAM, per its docs | The app plus MongoDB, Meilisearch, a RAG API and a vector database |
| From your phone | Your server, at an address you set up, with HTTPS | Your server, at an address you set up, with HTTPS |
| Updates and backups | Yours: pull new releases, back up the data | Yours: pull new releases, back up the data |
| License | Open WebUI License: BSD-3 terms plus a branding clause (since v0.6.6, April 2025): past 50 users the "Open WebUI" branding stays unless you buy an enterprise license | MIT |
| Paid options | Enterprise licenses through sales, for organizations only; no hosted plan | None listed; no hosted plan |
| Who's behind it | The Open WebUI project | ClickHouse, since November 2025; it says existing deployments continue and the open-source project keeps its investment |
Feature comparison
Checked October 2026 against each project's docs, repository and site. Both ship often; where a row says what's possible, it's from their own documentation.
| PolyCog | Open WebUI | LibreChat | |
|---|---|---|---|
| What it is | A websiteNothing to install; sessions in your account | An app you hostpip, Docker, Kubernetes or a desktop app | An app you hostDocker Compose, or Node.js with MongoDB |
| Software price | Free tier · $10/mo · $250 onceFree: 4 models, 3 debate rounds, 20 saved sessions | FreeIts own license; enterprise licenses by quote | FreeMIT license |
| The AI bill | Your keys, list priceBilled by your providers | Your keys, list priceOr a local model with no API bill | Your keys, list priceIn the server's .env, or each user's own |
| Also needed | NothingWe run the servers | A machine to run itPlus an address and HTTPS to reach it away from home | A machine to run itPlus MongoDB and three more services, bundled in Docker Compose |
| Several models at once | Debate + verify + synthesisParallel answers, difference scan, debate rounds, claims checked against web sources, one answer | Side by side + MergeCompare runs two or more models; Merge Responses (experimental) has one of them combine the drafts | Two at onceThe + button generates two responses side by side; no merge |
| Models | 6 providers · 13 AI labsOpenAI, Anthropic, Google, DeepSeek, xAI on their own keys, plus Qwen, Kimi, GLM, Mistral, Meta’s Llama and Muse, MiniMax, Venice and Nous through one OpenRouter key — current text models, added as released | Any OpenAI-compatible APIPlus Ollama and Anthropic connections | Most major providersOpenAI, Anthropic, Google, Azure, Bedrock, Vertex, plus any OpenAI-compatible endpoint |
| Local models | NoCloud APIs only | Yes, Ollama built inOne Docker image bundles Ollama | YesOllama through a custom endpoint |
| Tools and knowledge | Files + memoryUploads (pictures, PDFs, text) and a memory feature | Tools, RAG, arenaPython tools, MCP, 13 vector databases, web search, a model arena | Agents, code, RAGAgents, a code interpreter, artifacts, MCP, web search, memory |
| Spend controls | Estimate + monthly capIn dollars, rounded up; the cap stops runs | Token countsAdmin analytics count tokens, not dollars | Token balancesPer-user balances; costs shown once an admin turns it on |
| On your phone | Any browserSame account, same sessions | Through your serverAt the address you set up | Through your serverAt the address you set up |
| Upkeep | NoneUpdates ship without you | YoursPull new releases, back up the data | YoursPull new releases, back up the data |
| Try it first | Sample mode3 prompts a day on shared keys, no account | Install firstThen a key, or a local model | Install firstThen a key |
| Teams | Not yet | Multi-userRoles, groups, SSO, LDAP, SCIM | Multi-userSSO through OAuth, SAML or LDAP; an admin panel |
| Where your chats live | Your PolyCog accountSessions and encrypted keys on our server, so they follow you | Your serverOn-premises or air-gapped, if you like | Your server |
Open WebUI or LibreChat?
If you've decided to host your own, here's the short version of how they differ.
Pick Open WebUI if…
you want local models through Ollama built in, a knowledge base over 13 vector databases, Python tools and pipelines, a model arena with leaderboards, and side-by-side answers you can merge. Its own docs call it the broader platform.
Pick LibreChat if…
you want a focused multi-provider chat under the MIT license: agents, a sandboxed code interpreter, artifacts, MCP and memory, with sign-in through OAuth, SAML or LDAP. It runs two models side by side, with no merge.
Either way
both run on Docker, use your own API keys, keep your chats on your server, and make you the one who updates it. And in neither do the models read and answer each other before you get an answer. That step is what PolyCog is for.
Several models at once: side by side, or a debate
Open WebUI's Compare shows the answers side by side and can merge them; LibreChat shows two at once. PolyCog has the models answer each other first.
One prompt, two ways to ask several models: Open WebUI's Compare shows the answers side by side and its Merge has one model combine them; PolyCog runs the models in parallel, scans for disagreements, has them debate, and synthesizes one answer.
Where your chats live
Self-hosting puts your data on hardware you control. It also makes you the uptime.
That's the real trade. Open WebUI and LibreChat keep everything on your server, which is as private as it gets short of running the models yourself. PolyCog keeps sessions in your account, so they open on any device with nothing to run. Both are honest trade-offs, not flaws.
Where they beat us
A comparison you can trust has to cut both ways, and these are two very good projects.
Free, with the code open
Both cost nothing to run. LibreChat is MIT-licensed; Open WebUI uses its own license, free to use as long as the branding stays. PolyCog's free tier is free too, but Pro is $10 a month.
Local models, offline
Open WebUI has Ollama built in (one of its Docker images bundles it), and LibreChat reaches Ollama through a custom endpoint: models on your own hardware, no API bill, no internet. PolyCog runs cloud models only.
Your chats stay on your server
Chats, files and keys live on a machine you control, and Open WebUI lists on-premises and air-gapped setups. PolyCog keeps your sessions and encrypted keys on its server, which is what lets them follow you.
Tools, agents, knowledge
Open WebUI: Python tools and pipelines, MCP, a knowledge base over 13 vector databases. LibreChat: agents, a sandboxed code interpreter, artifacts, MCP, memory. PolyCog has file uploads and memory, and stops there.
Where PolyCog wins
Nothing to install, nothing to keep running
No Python, no Docker, no database, no releases to pull. PolyCog is a website: open it and ask. When something needs patching, we patch it.
A debate, not columns
Open WebUI puts several answers side by side and can have one model merge them; LibreChat shows two at once. In PolyCog each model reads the others and replies, a difference scan flags where they split, and a flagship writes one synthesis that names what's still disputed.
Claims checked against sources
A verify pass searches the web for each factual claim in the answers and shows the source and its exact words before the synthesis is written. Neither project documents a step like it.
On your phone, no server
Sessions live in your account, so a debate started on the laptop opens on the phone. A self-hosted app on your phone needs a machine that stays on, an address and HTTPS, all yours to set up.
Try it in ten seconds
Sample mode runs on our shared keys, three prompts a day, no account. The other two need an install and a key (or a local model) before the first word.
Common questions
Are Open WebUI and LibreChat free?
The software, yes. LibreChat is MIT-licensed; Open WebUI is free under its own license, which asks you to keep its branding once you pass 50 users unless you buy an enterprise license. What you pay for is everything around it: the API keys you plug in (the same list prices PolyCog users pay), a computer or server to run it on, and the time to install and update it. PolyCog's free tier costs nothing either, and there's nothing to run.
Open WebUI or LibreChat: which should I pick?
Open WebUI if you want local models through Ollama, a document knowledge base, Python tools and a model arena; its own docs call it the broader platform. LibreChat if you want a focused multi-provider chat with agents, a code interpreter and artifacts under the MIT license. Both run on Docker and both need you to keep them updated. If what you want is several models arguing a question out, neither does that; PolyCog does.
Can Open WebUI or LibreChat run several models at once?
Side by side, yes. Open WebUI's Compare sends one prompt to two or more models and shows the answers in columns, and an experimental Merge Responses button has one of those models combine the drafts. LibreChat can generate two responses at once with its + button. In neither does a model read another's answer and reply, which is PolyCog's core: a difference scan, debate rounds, a verify pass against web sources, then one synthesis.
Do I need a server to use them?
Not on your own computer: both install locally with Docker (Open WebUI also with pip). To use one from your phone or anywhere else, it has to run on a machine that stays on, at an address you set up with HTTPS, and you keep it patched. PolyCog is a website: the same account opens on any device.
Can I use the same API keys in PolyCog?
Yes. An OpenAI, Anthropic, Google, DeepSeek, xAI or OpenRouter key works in all three, at the same list prices. Plenty of people keep a self-hosted chat for everyday use and open PolyCog when a question deserves a debate.
Can PolyCog run local models?
No. PolyCog runs cloud models through their providers' APIs. If you need models that run on your own hardware, offline, Open WebUI with Ollama is the better tool.
See the difference in one prompt
Run the same question through several frontier models, watch them debate, and read the synthesis. Free sample mode, no signup, no keys needed.
Free tier forever · $10/mo · $250 lifetime · your API keys, your bill