Private AI agent workspaceThink freely.
Keep it yours.
Veilroom stores nothing. Open-weight models answer through providers that don’t keep or train on your data; GPT, Claude and Gemini are read by their host.
Your AI, wherever the work is.
Chat, create, hand work to agents and read the chain from one place, with the same privacy modes in every one of them.
- Sealed mode
- Ephemeral mode
- Never used for training
- Chat with any model
- Scheduled agents
- Web research
- Wallet watch
- Token checks
- Program reads
- Images and video
- Voice
- MCP tools
- API and SDKs
- Private vouchers
Tutorials and how it works, in under 20 seconds.
Short videos of a first chat, Sealed mode, agents, onchain lookups and top-ups. Every one uses sample data.
All guidesPick Sealed mode, ask in your own words, and read an answer only you can open, with a receipt of what was kept.
Hand off the busywork. Keep the strategy.
Set an agent once and it researches, watches wallets and reports back on a schedule, sealed to you. Pick a template and preview a run.
See every agentWeekly market brief
A one-page brief on what moved in your market, with sources.
- 1Search 38 sources
- 2Cross-check 112 claims
- 3Drop anything unsourced
- 4Write a one-page brief
The models you know, kept private.
Pick a model yourself or let Auto choose. Every model works in every mode: open-weight ones through private providers, the big hosted ones read by their lab.
- FreeAuto
Veilroom · Picks the right model for each question
Every mode · private provider
- For holdersLlama
Meta · Everyday chat and writing
Every mode · private provider
- For holdersDeepSeek
DeepSeek · Reasoning, maths and code
Every mode · private provider
- For holdersQwen
Alibaba · Many languages, long documents
Every mode · private provider
- For holdersKimi
Moonshot AI · Agents and tool use
Every mode · private provider
- For holdersgpt-oss
OpenAI · Open-weight reasoning
Every mode · private provider
- For holdersGPT
OpenAI · Hosted by OpenAI
Every mode · read by OpenAI
- For holdersClaude
Anthropic · Hosted by Anthropic
Every mode · read by Anthropic
- For holdersGemini
Google · Hosted by Google
Every mode · read by Google
Without the token you get Auto, which answers with a light open model. Holders choose any model and pay for use with credit. Every model works in every mode. Open-weight models answer through providers that don’t keep or train on your data; hosted models are read by the lab that hosts them. The launch lineup may still change.
- 01Privacy modes3Sealed, Ephemeral, Open
Pick per chat how much anyone else can see, from nobody at all to a standard provider.
- 02Models20In the preview lineup
Chat, image, video and voice, every one in every mode. Open-weight ones use private providers.
- 03Kept in Sealed0Prompts and answers
Encrypted on your device, opened only inside the enclave and by you. Nothing is stored.
- 04Minimum top-up$1USDC on Solana
On Solana, settled in seconds for under a cent, then turned into private vouchers that never point back to your chats.
The model count follows the preview lineup and will change as models are confirmed.
Every prompt says something about you.
Most AI apps keep what you ask, and some train on it. We don’t sell your prompts, we don’t train on them, and in Sealed mode we can’t read them at all.
Everything you’d ask an AI, kept private.
Chat, images, agents and onchain answers in one place, with the same privacy modes in every one of them.
- Chat
Ask anything, any model.
Quick facts to deep research. Auto picks the right model.
Auto · picked ReasoningsealedExplain my blood test in plain words. Anything to ask my doctor?bloodwork-june.pdf 3 pages6 medical sources checked webTwo values to raise, four look normal answer - Create
A sentence becomes a picture.
Images, short clips, voiceovers and music.
Create · 4 outputsImageClip · 6sVoiceoverEdit - Agents
Hand off the busywork.
Research, monitoring and reports that run privately too.
Agent · running privatelyEvery Monday · 08:00 brief38 sources read 2m 10sTell me when a rival ships watch - Onchain · built in
Read the chain in plain words.
Labeled, scored answers about any Solana wallet, token or program.
7xKp…Wq3F · 24h1 flagTagged: market maker 12 walletsFresh mint, freeze authority still onLiquidity moved to a stable pair - Developers
The same pipeline, in your app.
API, SDKs and MCP tools with the same modes.
await veilroom.chat({ mode: "sealed", model: "auto", messages, });
How a chat stays sealed, start to finish.
Sealed mode checks the hardware before you send anything, and opens the answer only on your device.
- 1
Check the enclave
Before anything leaves your device, your browser checks the enclave’s attestation, which proves exactly which code will handle your request.
- Attestation signed
- Code hash matches the published build
- 2
Encrypt on your side
Your prompt and files are encrypted to that enclave’s key, so they travel as noise to everyone in between.
What the network seeskQ7v2Lm9XbT0pW3eZr8cN1yHs4dJ6uFa5gVoE+/iB2tR9wKqY0nM3xC7lP8jUh= - 3
Work happens inside
The model, agent or onchain engine runs in isolated memory. Veilroom runs the machine but cannot look into it.
- Isolated memory
- No operator access
- 4
Only you open the answer
The answer comes back encrypted and is opened on your device. You get a receipt; the enclave keeps nothing.
- Opened on your device
- Nothing kept
Choose how much you reveal.
Set a mode per chat. Sealed is the default; switch when you want the widest choice of models.
- Default
Sealed
Nothing kept by VeilroomVeilroom stores nothing. Open-weight models answer through providers that don’t keep or train on your data; hosted models such as GPT, Claude and Gemini are read by their host to answer. Next: your prompt encrypted on your device to an attested enclave.
- Veilroom can read
- Model host can read
- Logs kept none
Best for contracts, financials, strategies and anything sensitive.
Ephemeral
Processed in memory, never storedRequests are processed in memory and discarded as soon as the answer is sent. Nothing is stored or used for analytics.
- Veilroom can read
- Model host can read
- Logs kept none
Best for everyday work where speed matters and nothing should be kept.
Open
Widest model choice, lowest costGives you the widest choice of hosted models. The provider reads the request to answer it and logs may be kept for a limited time.
- Veilroom can read
- Model host can read
- Logs kept none
Best for public information and casual questions.
Ask the chain
in plain words.
Veilroom reads Solana transactions and program logs and turns them into answers you can understand. Your questions stay private, so nobody sees what you’re watching.
- Wallets tagged as funds, market makers, MEV bots and dev wallets
- Risk flags on fresh mints with live mint or freeze authority
- Alerts the moment something moves
This wallet belongs to a market-making cluster of 12 addresses. It pulled 38% of its liquidity from one Raydium pool and moved it into a stable pair. Risk: low.
Built for anything you’d rather keep to yourself.
From deal notes to wallet research, Veilroom keeps the work that matters most out of shared logs.
Founders and operators
Draft, analyze and plan with contracts, numbers and investor notes that stay private.
Crypto teams and traders
Research Solana wallets and watch pools without broadcasting what you’re watching.
Developers
Ship AI features with the same private pipeline through the API and SDKs.
Legal and finance
Review documents and model scenarios without sending client data into shared logs.
Creators and marketers
Make visuals, scripts and voiceovers for launches you haven’t announced yet.
Students and everyone else
Ask questions, learn and brainstorm without your history becoming someone’s data.
Common
questions.
Still curious? Ask Veilroom itself, in a chat nobody else can read.
Ask in a private chatVeilroom doesn’t store your chats. Saved chats live only in your browser, and Ephemeral chats aren’t saved at all. Your message passes through Veilroom to the model: open-weight models use providers that don’t keep or train on it, and hosted models such as GPT, Claude and Gemini are read by their host. Sealed enclaves, where not even Veilroom could read a chat, are next.
Images, yes: holders can create them right in the chat. They’re made by Google’s Gemini image model, so Google reads the prompt. Video and voice are next.
Everyone can use Auto, which picks a light open model for each question. Holders can choose any model in any mode: open-weight ones such as Llama, DeepSeek, Qwen and Kimi, and hosted ones such as GPT, Claude and Gemini. The launch lineup may still change.
Not to start. Without a token you get Auto for free, up to 20 messages every 10 minutes. Hold any amount of $VEIL to unlock every model, or top up credit from $1 in ZEC and pay as you go.
No. Chat and Create work for any kind of question. Onchain answers are there when you need them.
No. Nothing you send through Veilroom is used to train models, in any mode.
Think freely.
Keep it yours.
Chat, create, hand off research and read the chain in one workspace that keeps no copy of what you ask.
A little room. A bigger idea.
An original 18-second introduction to the proposed workspace.
Access, with clear boundaries.
A proposed utility token for model access. No token has been issued. All figures below are design proposals inherited from the source, not live chain data.
Proposed supply · 999,928,962 VEIL
Proposed: six decimals, revoked mint and freeze authorities, Token-2022 metadata extensions. Allocation, tax, burn, dividends, circulation and locking are unknown and await a separate proposal. Holding any amount would unlock model choice; Auto would remain free.
Start free. Choose your way.
Proposed access follows the source: Auto at no cost, 20 messages per 10 minutes; holders unlock model selection. Usage credit starts at $1. Provider cost plus 10% is a source pricing mechanism, not an active Veilroom offer.