The world's most powerful open source models.
The open weights ChatGPT and Claude alternative.
100% Private.
Multi-layered memory.
A full workspace and voice.
GLM 5.2, Kimi 2.7, MiniMax M3,
DeepSeek V4, Nemotron 3, MiMo, and more.
Hosted on US infrastructure. 100% private and secure.
NewBank up to a week's worth of usage and use it on busy days.
The world's most powerful open source models are here. They rival or beat the big AI models. But actually using them privately, with real memory, real voice, skills, websearch, and real tools? Until now, that meant learning Docker, buying an expensive GPU, or sending your data to servers you don't control.
Open Grove changes that. One subscription gives you direct access to frontier open source models running on private US infrastructure, with the dynamic memory, voice, and bias mitigation architecture that Phoenix Grove Systems is known for. No data harvesting. No training on your conversations. No catch.
Making Chinese AI safe, accessible and fully private. Open source AI, without the setup.
Use any model. Keep your privacy.
Your prompts never touch the original model developer's servers. Your conversations never become training data. Privacy is a right, one you deserve. One you should have without buying expensive GPUs or local setup.
Your data belongs to you.
Is Chinese AI safe?
Yes, entirely, when it runs on private US servers and your data never goes to the original labs. The concern was never really the models. DeepSeek, GLM, and Kimi are open weight models: published files anyone can inspect and host. A model file cannot send your data anywhere. Where your conversations go is decided by whose servers run the model.
Use a Chinese company's official app, and your conversations travel overseas under another jurisdiction's laws. Use the same open weight models on Open Grove, and every prompt is processed start to finish on private US infrastructure: nothing routed to the original labs, nothing retained for training, no telemetry. The intelligence stays frontier. The data stays home.
That is the entire safety question, answered by architecture instead of trust. It is why Open Grove exists: making Chinese AI safe, accessible and fully private, and making open source AI usable without the setup.
What makes Open Grove different.
Every model runs with the full PGS platform behind it. Here's what that means.
Memory that actually remembers.
Open Grove remembers your projects, your preferences, your context, across conversations, across days, across models. Switch from GLM to DeepSeek mid-thought and your AI still knows everything that matters.
Every piece of memory is yours to view, edit, or delete. You're never starting from scratch again. Not with any model.
Voice that doesn't downgrade.
Switch to voice and the model stays the same. DeepSeek V4 Pro doesn't become DeepSeek V4 Flash. The intelligence, the memory, the context stay intact. Tap to interrupt anytime.
Intelligence, not ideology.
Every model runs under our seven-principle Ethical Charter with structural bias interventions. Not censorship. Just sycophancy reduction, stereotype interruption, and honest output.
Watch the thinking happen.
Reasoning traces stream in real time so you see exactly how the model is approaching your question. No black box. Full transparency into the cognitive process.
New models within 48 hours.
When the frontier ships, Open Grove keeps up. No platform switching, no new APIs, no waiting for product teams to evaluate. The newest intelligence, ready when you are.
Curated frontier intelligence.
Not every open source model makes the cut. We select for quality, capability, and models that meet our standards for responsible deployment. Each model carries a usage band, Everyday, Advanced, or Frontier: the same bands the plans below use to describe your day.
This lineup grows with the frontier. When a major new model releases, it's typically available here within 48 hours.
Prefer to run them on your own hardware? We keep a free community guide to the best quants, licenses, and honest hardware requirements for every model here: the Open Model Quant Guide.
Bring all of your chats with you.
Your entire ChatGPT, Claude, or Gemini history moves into Open Grove as your actual chats: same conversations, original dates, right in your sidebar. And they become living memory, so you can ask about a project from two years ago and just keep going, with any model we host. Request your official export, drop the file in, done in minutes.
Your chats belong to you, not to a company.
Step 1: Get your export from ChatGPT or Claude
Every provider is required to give you your data. Here is where the button hides:
- ChatGPT: click your name → Settings → Data controls → Export data.
- Claude: click your name → Settings → Privacy → Export data.
- Confirm the export. Within about 10 to 15 minutes you will get an email with a download link.
- Download, unzip, and find the file named conversations.json in the main folder. That single file is your entire history.
Heads up: on ChatGPT the Export data button sits right below Delete data. We know, it stresses us out too. Click carefully, and do not click delete.
Step 1 for Gemini: get your Google Takeout file
Google routes it through Takeout, so it is a few more clicks:
- Go to takeout.google.com and click Deselect all.
- Scroll to My Activity and check its box.
- Click Multiple formats and change Activity records to JSON.
- Click All activity data included and select only Gemini Apps.
- Next step → Create export, then wait for Google's email (minutes to hours).
- Download, unzip, and find MyActivity.json. That is your file.
Google structures its history differently, so Gemini chats import rebuilt by timeline rather than as exact conversations. Still your history, still remembered.
Step 2: Import with Memory Forge
- In PGS AI, open Settings → Import Your Chats with Memory Forge
- Choose your export file. You can pick several at once, even from different services.
- Review the list of conversations it found and pick where they should live. A fresh AI Node is recommended, so this history gets its own mind.
- Start the import. Your chats are created first, then woven into your AI's memory. ChatGPT and Claude arrive as your exact chats with their original dates and an import badge.
Big archives can take a while, and that is fine: if anything interrupts the import it picks back up where it left off, nothing lost, nothing duplicated. The memory weaving finishes on our side, no tab required. Then just keep talking, with DeepSeek, GLM, Kimi, whichever mind you like.
Prefer not to move in? Make a free memory chip instead
Memory Forge can also condense your history into a portable memory chip file. Chip files are generated entirely in your browser: zero network activity, the data never touches our servers, and you can verify that yourself with your browser's developer tools. Take the chip to any AI service that will read it, including away from us, forever.
A door that only swings inward is just lock in with better marketing. Either way, we've got you.
Your subscription is also a coding plan.
Every paid plan includes an API key for your coding tools and agents. Near-endless messaging on these same open source models, native support for the popular coding tools, and unused time that banks for your heavy weeks. It has its own home now, with pricing, docs, and per-token options too.
Open Grove is one half of something larger.
The same subscription that gives you Open Grove also unlocks Phoenix Grove, our multi-core AI platform where we build something nobody else is attempting. Parallel cognitive architectures where multiple reasoning engines fire at once and synthesize into responses that no single model can produce. Visible cognition. AI builds with genuine depth and personality.
One toggle in the app. Same memory. Same voice. Open Grove gives you the world's best open source models, privately. Phoenix Grove shows you what's possible when you stop treating AI as a model and start treating it as a mind.
Open Grove + Phoenix Grove. One price.
Every tier unlocks both Open Grove's open source lineup and Phoenix Grove's multi-core builds. Climb the ladder for more models, more Grove builds, and a higher ceiling. Start with a free month, on us.
On usage: every plan gives you one flexible daily allowance shared across all models, and lighter models stretch it further. We group the lineup into three bands by how much a message draws:
- GLM 4.7 Flash
- DeepSeek V4 Flash
- Gemma 4 (31B)
- MiMo V2.5
- MiniMax M2.7
- GLM 5
- GLM 5.1
- MiniMax M3
- Kimi K2.6
- Kimi K2.7
- Nemotron 3 Ultra
- GLM 5.2
- DeepSeek V4 Pro
- MiMo V2.5 Pro
Each tier below shows roughly what a day looks like in each band. Numbers vary based upon usage patterns, but these are good estimates for most people.
Your unused week stays yours.
Most services rely on you letting usage go to waste. We decided to take a different path, and allow users to bank unclaimed usage. Usage banking is automatic, and any unclaimed usage at the end of each week will be added to your usage bank (up to one week of usage banked at a time). On days or weeks where you need more usage, it's there waiting for you. Banking eases the stress of usage windows, so you can relax and use the service when you need it.
Daily estimates reflect typical conversations. Long documents, deep research, and very long conversations use more; lighter models stretch your allowance further, and you can switch models anytime, even mid conversation.
Every tier includes full privacy, full memory, full voice, and tools. The essentials never sit behind a paywall. Upgrade, downgrade, or cancel anytime.
Quick answers
Is Chinese AI safe to use?
Yes, when it is hosted correctly. Models like DeepSeek, GLM, and Kimi are open weights: files with no network access of their own. The safety question is entirely about whose servers run them. On Open Grove they run on private US infrastructure, your prompts never reach the original labs, nothing is retained for training, and there is no telemetry. Same intelligence, none of the data exposure.
Can I use open source AI without downloading or setting anything up?
Yes. That is what Open Grove is: the leading open source models, ready in your browser. No GPU, no local install, no API keys, no Docker. One subscription, full memory, voice, and workspace, and a free trial month to see how it feels.
How is this different from using DeepSeek or GLM directly?
When you use a model on its developer's own servers, your data is processed on their infrastructure under their jurisdiction's laws. Open Grove runs the same models on privacy-first US infrastructure. 100% private. Your prompts never touch the original developer's servers.
Beyond privacy, you get persistent memory across all conversations, full voice mode without model downgrades, bias mitigation, web research, file uploads, and a code canvas. The model is the same. Everything around it is better.
What does "bias mitigated" mean?
Every model running through Open Grove operates under Phoenix Grove Systems' Ethical Charter and structural bias interventions. These reduce sycophancy (the AI agreeing with you when it shouldn't), interrupt common stereotype patterns, and resist confirmation bias.
This is not censorship. The models retain their full capability. The interventions help them respond with genuine honesty and nuance rather than telling you what you want to hear or reproducing biases from their training data.
Do you train AI models on my conversations?
No. Not now, not ever. Unlike most AI companies, we believe deeply in data sovereignty, privacy and basic respect. We can say honestly that standard data practices in the AI industry are deeply disturbing to us, and should be to everyone. To us, respecting our user's privacy is a matter of common decency. We have built architecture that promises: Your data is yours, and will stay that way.
What's the difference between Open Grove and Phoenix Grove?
Open Grove gives you direct access to frontier open source models with memory, voice, and privacy. You're talking to GLM, or Kimi, or DeepSeek, enhanced by PGS infrastructure.
Phoenix Grove is our multi-core cognitive architecture: multiple specialized reasoning engines running in parallel, synthesized through an executive layer. These are bespoke AI builds with visible cognition, distinct cognitive personalities, and architectural depth that goes far beyond a single model.
Both are included in the same subscription. One toggle switches between them. Your memory carries across both.
How quickly do you add new models?
Typically within 48 hours of a major frontier release. Not every model makes the cut. We select for quality and responsible deployment, and we exclude models with known bias issues or quality problems that don't meet our standards.
Can I switch between models mid-conversation?
Yes. Your memory and conversation context carry across models. Start a project with GLM 5.2, switch to DeepSeek V4 Pro for a different perspective, move to Kimi K2.7 for code work. The AI remembers everything regardless of which model is active.
Why are some models excluded?
We curate for quality and responsibility. Some models have documented bias issues that our mitigation layer cannot adequately address. Others don't meet the quality threshold we set for what carries the Open Grove name. We'd rather offer fewer models we're proud of than pad the list with names that disappoint.
The world's best open source models. Private. With memory, workspace and voice.
One subscription. Full privacy. Full voice. Try free for a month and see if it fits.