Claude alternatives: 8 free & open-source picks
Compare recorded free and open-source options, their capabilities and operating requirements. Source dates are shown below; licenses, hosted limits and project activity can change.
A private AI desk for documents and agents; install it, then choose local models or somebody else's meter.
licenseMIT
runsyour machine or their cloud
installNative desktop installer, or Docker for a shared/self-hosted server
enginesLocal llama.cpp and imported GGUF models, plus Ollama, LM Studio, LocalAI and numerous cloud providers including OpenAI, Anthropic, Google, Azure, AWS Bedrock, Groq, Mistral, Cohere, and OpenRouter.
dataDesktop data stays in the app's local data directory; the Docker edition persists its server storage volume, including an SQLite database and the default LanceDB vector store. Configured cloud providers receive the prompts and files sent to them.
the catch vs ClaudeIt does not include Anthropic inference, so matching Claude's reasoning and document quality requires a separately metered Anthropic API key or a usually weaker local model.
setupA non-expert must choose a model path: download a local model with enough RAM, connect Ollama or LM Studio correctly, or bring and pay for an API key.
facts verified 2026-08-10
A desktop model switchboard with assistants, files, agents and MCP; bring keys or make your laptop sweat.
licenseAGPL-3.0
runson your machine
installNative macOS, Windows, or Linux desktop installer
enginesSupports hundreds of models through providers such as OpenAI, Anthropic, Google Gemini, OpenRouter, Azure, AWS Bedrock, Groq, DeepSeek, Mistral, Ollama, LM Studio, and other OpenAI-compatible services.
dataChats, API keys, assistants, settings, and local knowledge-base data are kept on the device; optional backup can use local files or WebDAV, while configured model providers receive prompts and attachments.
the catch vs ClaudeIt does not bundle Claude usage, and its desktop-first workspace is less seamless for Claude's first-party document, project, and cross-device workflow.
setupThe installer is easy, but useful output requires either a paid provider key or downloading and running a local model that fits the machine.
facts verified 2026-08-10
A local-first desktop assistant with cloud connectors and tools; your RAM gets the invoice.
licenseApache-2.0
runson your machine
installNative macOS, Windows, or Linux desktop installer
enginesRuns local GGUF models through llama.cpp and Apple-silicon models through MLX, and connects to OpenAI, Anthropic, Google Gemini, Groq, Mistral, Azure, OpenRouter, Hugging Face, and custom OpenAI-compatible APIs.
dataModels, assistants, files, and settings stay in Jan's data folder; threads are stored as local JSON and JSONL files. Cloud providers receive only the conversations or files routed to them.
the catch vs ClaudeThe free local route usually gives up Claude's reasoning and long-document quality, while connecting Anthropic turns the replacement back into metered API spend.
setupThe first choice is consequential: local models need a sizable download and enough RAM, while cloud models require separately billed API keys.
facts verified 2026-08-10
A self-hosted model switchboard with agents, files and tools; free software, paid model bills if you use paid models.
licenseMIT
runsyour server
installOfficial Docker Compose stack
enginesConnects to OpenAI, Anthropic, Google and Vertex AI, Azure OpenAI, AWS Bedrock, Mistral, Groq, custom OpenAI-compatible endpoints, and local backends exposed through compatible APIs.
dataThe default stack stores conversations and accounts in MongoDB, search indexes in Meilisearch, and uploaded or RAG files in local volumes or configured object/vector storage; every configured model provider receives the requests routed to it.
the catch vs ClaudeLibreChat can call Anthropic but does not include it, and it replaces Claude's managed service with a MongoDB-based system the user must secure and maintain.
setupIt needs Docker, environment secrets, provider keys, storage, backups, and an admin; the official guide also documents a MongoDB image override because the default image crashes on Apple-silicon Macs.
facts verified 2026-08-10
A local-first multi-model workspace with files, search, code and personas; the interface assumes you enjoy buttons.
licenseMIT
runsyour machine or their cloud
installUse the hosted web app, or deploy the Next.js app yourself
enginesConnects to more than twenty backends, including OpenAI, Anthropic, Google Gemini, OpenRouter, Azure OpenAI, AWS Bedrock, Groq, Mistral, DeepSeek, Ollama, LM Studio, LocalAI, and OpenAI-compatible endpoints.
dataIn the browser edition, provider keys and settings are stored in localStorage and chats, files, personas, and related workspace data in IndexedDB; configured model providers receive submitted content. There is no included cross-device cloud sync.
the catch vs ClaudeBig-AGI offers a multi-model shell rather than Claude's included Anthropic models, first-party Projects and Artifacts workflow, and managed synchronization.
setupThe first obstacle is bringing provider keys or a local endpoint; the app warns that browser-stored keys are plain text and protected only by the browser profile.
facts verified 2026-08-10
A local-model desktop lab with chat, documents and tools; free forever, but your hardware is now the cloud.
licenseproprietary, free
runson your machine
installNative desktop installer or Linux AppImage
enginesRuns downloaded open-weight GGUF and safetensors models locally through llama.cpp and MLX-compatible runtimes, and exposes local OpenAI-compatible APIs.
dataDownloaded models live in LM Studio's local model directory and chats are stored as local JSON conversation files; local inference does not send prompts to a model provider.
the catch vs ClaudeLocal open-weight models usually fall short of Claude's strongest reasoning, writing, and long-document performance, and LM Studio lacks Claude's managed project workflow.
setupThe first obstacle is hardware: LM Studio recommends 16 GB RAM, requires current supported operating systems and CPU features, and useful models can occupy many gigabytes.
facts verified 2026-08-10
ChatGPT-shaped self-hosting with models, files, search and tools; the software is free, the admin role is yours.
licenseOpen WebUI License
runsyour server
installDocker is the recommended path; pip, uv, Kubernetes, and desktop-packaged paths are also documented
enginesConnects natively to Ollama and OpenAI-compatible APIs, with documented support for providers and gateways such as OpenAI, Anthropic, Google, Azure, Groq, Mistral, OpenRouter, and local inference servers.
dataA default single-node install persists accounts and chats in webui.db (SQLite), embeddings in Chroma, and uploads in the Open WebUI data volume; larger deployments can move these to PostgreSQL, Redis, and external vector or object stores.
the catch vs ClaudeIt can front Anthropic-compatible access but supplies neither Claude usage nor Anthropic's first-party Projects and Artifacts experience, while adding server administration.
setupThe first obstacle is owning the service: Docker volumes, a model backend or API key, user security, updates, and backups; an unpersisted container loses the data.
facts verified 2026-08-10
A polished agents-and-files workspace; self-hosting is free and frontier inference usually is not.
licenseLobeHub Community License
runsyour machine or their cloud
installHosted signup, one-click Vercel deployment, or Docker Compose for the server-database edition
enginesSupports more than seventy providers and gateways, including OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, OpenRouter, Groq, DeepSeek, Ollama, LM Studio, and OpenAI-compatible endpoints.
dataLocal mode keeps chats in browser IndexedDB and supports JSON export; server mode uses PostgreSQL for account and conversation data plus S3-compatible object storage for files. Configured providers receive routed prompts and attachments.
the catch vs ClaudeUsing Claude through LobeHub still requires separately billed Anthropic access, and self-hosted server mode is materially more complex than Claude's managed workspace.
setupHosted use still requires model credentials or purchased inference, while serious self-hosting adds PostgreSQL, object storage, authentication, secrets, a domain, and migration from browser-local data.
facts verified 2026-08-10
latest recorded source check 2026-08-10 · check current terms at the source
MstyThe desktop core is $0, but the terms limit it to private non-commercial use; paid work needs a license.
| tool | license | running it | platforms | stars | active |
|---|---|---|---|---|---|
| AnythingLLM | MIT | one-click install | web, macos, windows, linux, self-hosted | 64,400 | 2026-08 |
| Cherry Studio | AGPL-3.0 | one-click install | macos, windows, linux | 49,867 | 2026-08 |
| Jan | Apache-2.0 | one-click install | macos, windows, linux | 43,900 | 2026-08 |
| LibreChat | MIT | docker to self-host | web, self-hosted | 41,700 | 2026-08 |
| Big-AGI | MIT | docker to self-host | web, self-hosted | 7,100 | 2026-08 |
| LM Studio | proprietary, free | one-click install | macos, windows, linux | — | — |
| Open WebUI | Open WebUI License | one-click install | web, macos, windows, linux, self-hosted | — | — |
| LobeHub | LobeHub Community License | one-click install | web, self-hosted | — | — |
They pay because the quality sits in the model and product platform, not the wrapper. The recorded subscription totals $240 a year before any costs of an alternative.
Questions about these alternatives
Is there a free alternative to Claude?
Yes: AnythingLLM, Cherry Studio, Jan and 5 more. These are recorded free or open-source options. Compare their feature limits, licenses, hosting costs and current availability before choosing.
Should I just build my own Claude?
Our verdict is NOT REALLY. A Claude-like UI is trivial; the model, safety/reliability layer, long-context performance, artifacts/projects, and platform distribution are not. Review the scope, implementation prompt and trade-offs in the Claude build guide.
