Aura FHE Chat
View the Chat page, then contact Gen to arrange a live demonstration or discuss access. Bring a useful question and a public or invented example.
Explore the demonstration ↗AURA / ENCRYPTED COMPUTE
Your data.
Your terms.
Bringing FHE into real-world AI.
Your data should work for you. On your terms. Aura is building decentralised AI with encrypted computation, starting with FHE Chat.
FHE Chat can be demonstrated now. Other applications are in development.
ENCRYPTED IN.
ENCRYPTED OUT.Conceptual artwork
ENCRYPTED OUT.Conceptual artwork
START HERE / FROM IDEA TO APPLICATION
Explore the application.
Help shape what comes next.
FHE means computing with encrypted data. FHE Chat is the application we can demonstrate now. The wider platform is in development, with a separate public learning preview for developers.
More applications
Encrypted databases, messaging and other applications are being built. Tell us which problem you want to solve and what would make the result useful.
Discuss an application idea ↗Aura MCP
Explore supported numeric operations with fixed public examples and Aura-managed demo keys. This developer preview is separate from FHE Chat.
Explore the learning preview ↓FHE Chat can be demonstrated now. View chat.afhe.io and request a live demo through Gen at gen@afhe.io. Most demo and access requests go through Gen. Viewing the page does not grant unrestricted access. Existing invitees retain their access. Start with public or invented examples.
PUBLIC MCP / CONNECT YOUR ASSISTANT
Learn the foundation.
Build with understanding.
MCP is a standard connection between assistants and external tools. Aura MCP helps you understand FHE, try supported encrypted computations and plan how those operations fit your application.
NPM / PUBLIC PREVIEW
Install. Connect. Ask.
Install Node.js 20 or later. Run these commands on the computer that runs your assistant, then paste the generated configuration into its MCP settings and reconnect. No Git, npm account or demo keys are needed.
aura_startExplore Aura AI, learn FHE and plan an application integration.fhe_status · fhe_opsCheck the numeric service and discover supported operations.fhe_inputs · fhe_computePrepare public encrypted samples and request a calculation.fhe_export · aura_proofSave encrypted results and understand what the evidence supports.
Connect Aura MCPnpm package ↗
npm install -g @aurafhe/mcp@preview
aura-fhe-mcp --config cursor --demoRun locally, then paste the generated settings.
YOUR FIRST MESSAGE
“What can I do with Aura? Show me the AI application first, then explain what I can run here.” Or ask: “Teach me how FHE works, then help me plan an application using the supported operations.”
Application walkthrough ↗See the whole flow.
Follow encrypted inputs through computation and result storage. Learn why secret keys determine who can read the result. Then try a public average or weighted calculation.
Map your application.
Choose the task and supported operations. Define the client, key storage, authenticated compute service and authorized result recipient. FHE becomes part of that application workflow.
Build toward more.
Explore analytics and scoring with the numeric tools. Discuss encrypted databases, messaging and AI through their separate integrations. Each use case needs its own design and evaluation.
Simple demo setup; a fuller path for private applications. The public lesson needs package installation and host configuration. Confidential applications also need local encryption and decryption, keys, authentication and output permissions. Browser WASM integration is planned; this MCP does not yet package a one-click confidential workflow. Learning and integration guide ↗
AI should do more.
You should give up less.
An assistant that helps with work or daily life needs context to be useful. Aura is building towards AI that can use sensitive information while people and organisations retain control over who can read it. FHE is the foundation; decentralisation is the wider network direction.
- 01
Assistants
Answer our questions.
- 02
Agents
Act across our tools.
- 03
Transactions
Execute our decisions.
- 04
Organisations
Coordinate our systems.
- 05
Machines
Learn from our world.
THE AURA THESIS
Make data useful.
Keep ownership personal.
The goal is to let an AI system use sensitive information without handing readable records to the computers doing the work. Encrypted computation is the foundation; an open network is how Aura intends to make it widely available.
Privacy becomes
part of everyday life.
An assistant working with personal records. A market that keeps bids confidential. A robot learning without exposing what it sees. FHE Chat can be demonstrated now; the other application areas below are in development.
FHE CHAT / AVAILABLE TO DEMONSTRATE
A useful question. An encrypted foundation.
Explore Aura FHE Chat with a live demonstration. Bring a question and a public or invented example, then discuss the application experience and the data boundary with the team.
FHE Chat can be demonstrated now. View chat.afhe.io and contact Gen at gen@afhe.io for demonstrations or access. Other applications are in development. The public numeric MCP learning preview is separate.
ILLUSTRATIVE APPLICATION / TARGET FLOW
- 01An invented business memo
Public or fictional content for the first demonstration
- 02An AI task in the separate application
Explain the actual encryption and key-custody path
- 03A summary and follow-up answer
Show the response and observed timing, if measured
Private at the source.
Distributed by design.
Three parts work together: encryption protects the data, MCP connects AI assistants to tools, and blockchain coordinates payments and participation. The architecture below shows the network Aura is working toward.
OWNER-CONTROLLED / TARGET ARCHITECTUREWho can decrypt the data? ↘
EXPLORE THE TARGET ARCHITECTURE
01 / YOUR DEVICEThe key stays here.
02 / AURA COMPUTEOnly encrypted work.
03 / BLOCKCHAINCoordinate. Verify. Settle.
ILLUSTRATIVE NETWORK · NOT LIVE ACTIVITY
The intended owner-controlled model: encrypt on the device and retain the key.
Target architecture · see current MCP boundaries ↗Privacy begins
at the source.
In the target design, data is encrypted within your device or trusted environment. Your keys stay there. Only an authorised recipient can decrypt the result.
MCP connects.
FHE protects.
An AI assistant requests a supported operation through MCP. Fully homomorphic encryption (FHE) lets network computers process the data while it stays encrypted.
A network with
work to do.
Blockchain provides a shared record of payments and participation. Independent operators can contribute compute, with their work checked before they earn fees. This is the planned open-network model.
BLOCKCHAIN COORDINATIONNetwork economics ↗
Solana first. An open network next.
The launch plan uses Solana to record and settle payments. AURA pays for compute; SOL covers the transaction fee on Solana. Independent operators join after the required review and testing.
Whitepaper v5.1 starts with compute operated by the Aura Foundation. Independent operators join after review and reproduction of the results by third parties. The current MCP integration has different privacy limits: read them before connecting ↗.
Send the work.
Keep the key.
Fully homomorphic encryption (FHE) lets a computer calculate without first decrypting the data. Aura’s target design keeps encryption, keys, and authorised decryption within the owner’s device or trusted environment. Compute operators receive no decryption key.
Owner’s device→Encrypted computation→Owner’s device
Seal
Create and keep the keys within your own trusted environment. Encrypt the data before sending it out.
Compute
Run supported operations while the inputs, intermediate steps, and results remain encrypted.
Reveal
Return the encrypted result to its authorised recipient. Decrypt it within the environment that holds the keys.
Owner-controlled custody
Whitepaper v5.1 requires approval from a defined number of key holders within the owner’s environment, so one person cannot decrypt alone. Aura operators and developers have no decryption role or recovery key in this model.
Approval by a defined committee
Applications combining data from several owners, such as sealed auctions, use a different model. A defined number of committee members must authorise decryption. Membership and rules must be disclosed; compute operators receive no decryption-key shares.
Precise privacy, defined scope. FHE does not by itself hide which operations run, when they run, or network activity. It also does not hide information you choose to reveal to an AI model. Device security and control of the keys still matter.
What makes the ZX-Engine different?
The whitepaper describes a lookup-based FHE construction designed to avoid repeated bootstrapping within Aura’s supported execution architecture. It reports deterministic, bit-exact outputs for redundant verification and a 3.7 MB edge runtime.
The scheme’s security assumptions, production parameters, and post-quantum standing remain subject to independent review. The runtime and benchmark claims are reported by Aura; the public MCP repository is not a publication of the cryptographic engine.
Review the evidence and its limits ↗An economy powered
by useful work.
In the proposed network, people pay for encrypted computation and operators earn fees for completing it. Proof of Encrypted Work (PoEW) is the planned method for checking that work was delivered. Solana records and settles the payments.
- 01
Request
A user submits a supported encrypted job.
- 02
Assign
The protocol assigns work to independent nodes.
- 03
Verify
Several nodes repeat the computation and compare their results.
- 04
Settle
Verified work earns fees. Results that do not match enter a dispute process.
TOKEN DESIGN / WHITEPAPER v5.1
AURA pays for the work.
Compute fees are designed to reflect workload complexity, hardware cost, and verification overhead. The token pays for network services and, after the required review, supports operator deposits (staking) and rewards.
SOL covers Solana transaction fees. Enterprise deployments generate AURA network fees only when they use the public network.
Participation follows readiness.
The network is in development. Operator requirements, verification rules and network economics need a consistent published specification before participation opens.
One destination.
Aura L1.
A future blockchain built around encrypted computation.
A layer-one blockchain (L1) runs its own rules for agreeing on and recording transactions. Aura plans to start on Solana, open computation to independent operators, then build its own chain when the technical and economic requirements are met.
Launch on
Solana.
Launch the token and settle payments on Solana. Begin with compute operated by the Aura Foundation.
An open
compute network.
Independent operators deliver encrypted work. The protocol verifies service delivery.
Aura L1.
A dedicated chain for encrypted work, built once the network meets its readiness requirements.
Future milestone · no fixed launch date.Every step has to be earned.
FHE Chat demonstrations and the public developer learning preview are the current starting points. Other applications and independent network participation are in development. Expansion follows evidence and readiness.
01CURRENT PRIORITY / FHE CHATBring FHE into real-world applicationsChat demonstrations now
Demonstrate FHE Chat, learn from useful questions and build the next applications. Developers can separately explore supported numeric operations in the public MCP learning preview.
Other applications are in development. Demonstration availability does not establish a production-readiness or security certification.
02PLANNED COORDINATION LAYERDevelop payments and coordination on SolanaDevelopment milestone
Solana is the planned initial payments and coordination layer. The wider goal is a network for encrypted computation. Token arrangements and participation requirements will be communicated through a consistent published specification.
No fixed launch date is announced here. Independent operator participation remains in development.
03HARD GATEPublish the evidenceReview and reproduction
Publish technical materials, independent review and reproducible results with the relevant workload, release and data/key boundary. Progress depends on what that evidence supports.
Independent operator participation requires technical review, reproduction and published joining requirements.
04AFTER REVIEW & REPLICATIONWelcome independent operatorsAfter review and testing
Onboard independent operators under published hardware and staking rules. Then activate operator deposits, penalties for breaking the rules, and PoEW rewards. Foundation-operated clusters retire progressively.
Participation, assignment, verification, and appeals require published rules.
05PLANNED EXPANSIONReach more ecosystemsFuture development
Explore connections to other blockchains, enterprise applications and physical AI as technical readiness and resources allow. These are development directions, not available products.
The initial focus is learning and memory that can tolerate a network delay. Immediate sensing and safety-critical control stay on the device.
06SIX CONDITIONS / NO FIXED DATEThe sovereign Aura L1When the network is ready
Aura L1 requires six things: published technical review; sustained network fees; demand from other blockchains; a mature method for verifying compute; a dedicated blockchain-consensus team; and security experience from production use.
The decision depends on evidence from a working network.
From cryptography
to useful computation.
The ZX-Engine is Aura’s encrypted-compute engine. These internal test results from whitepaper v5.1 cover AI models, database queries, and specialised hardware. They still require independent review and reproduction.
20B encrypted inference
GPT-OSS-20B, a 20-billion-parameter AI model, on one RTX PRO 6000 Blackwell GPU. Reported generation throughput; prompt read-in below 2.3 seconds. Hardware, workload and timing scope belong to this internal benchmark, not every visitor session.
Encrypted database query
An indexed query across 1,000,202 rows in the RainDB test database. This timing does not apply to every database query.
Encrypted logic throughput
Per Encryption Processing Unit (EPU), Aura’s specialised accelerator. Logic operations per second measure hardware throughput, not AI model speed.
Public integration code
Review the available MCP tools, how they connect, and where encryption and decryption happen.
Open Aura MCP ↗Technical evaluation
Request the whitepaper and evaluation materials: test conditions, hardware, security parameters, and access to reproduction tools.
Request technical materials ↗Independent review
Independent review and reproduction of the results are required before opening operator participation. The reviewed materials do not yet include published independent validation.
See the verification sequence ↗Numeric correctness checks, AI throughput benchmarks and security review answer different questions. A successful arithmetic result does not establish model performance or key custody. Sources: Aura public whitepaper v5.1, September 2026, §§03–14; official MCP public preview, reviewed 24 September 2026. Request benchmark scope and detailed methodology through the technical evaluation process. These figures are not production guarantees or live MCP inference capabilities.
AURA INSIGHTS / FROM THE FOUNDER
Ideas for a more
private AI future.
Why control over your information belongs at the centre of useful AI, and what it takes to bring encrypted computation into everyday applications.
Explore all articles ↗01 / THE AURA THESIS
The most useful AI will need the data we hesitate to share
The next step for AI should let people bring more of their world into a conversation while retaining control over it.
Read the article ↗02 / UNDERSTANDING ENCRYPTED COMPUTATION
What FHE changes about who gets to use your data
The ability to perform a calculation can be separated from the ability to read the information inside it. That changes how we can design services.
Read the article ↗03 / BUILDING FOR REAL-WORLD USE
The work between an FHE demo and a dependable product
Production is an experience people can rely on. Getting there means paying attention to the whole task, from the first input to a useful result.
Read the article ↗Clear by design.
Talk with the team ↗What is Aura?
Aura is building decentralised AI with encrypted computation. Our immediate focus is bringing fully homomorphic encryption into real-world AI, starting with FHE Chat.
What does FHE mean?
Fully homomorphic encryption enables supported calculations on encrypted data. The application, encryption boundary, key custody and result access determine the protection of a complete workflow.
What can I explore today?
FHE Chat can be demonstrated now. View the Chat page and contact Gen for a live demonstration. Other applications are in development. Developers can separately explore the public MCP learning preview.
How do I request a demonstration?
Email gen@afhe.io with what you would like to explore, your preferred time and timezone, and your organisation if relevant. Most demo and access requests go through Gen. A request does not automatically confirm a booking. Read the demonstration guide.
Does viewing the Chat page give me access?
Visitors may view chat.afhe.io. Viewing the page does not grant unrestricted product access; most demo and access requests go through Gen. Existing invited users retain their access.
Is Aura only a Solana project?
Aura’s wider direction is decentralised AI with encrypted computation. Solana is the planned initial payments and coordination layer. Independent network participation and a future Aura L1 remain development directions.
Is the public MCP the same as FHE Chat?
No. The public MCP is a separate numeric learning preview using fixed public examples and Aura-managed demo keys. It does not establish production confidentiality or encrypt an existing host assistant. Anything entered into a host assistant is visible to that assistant provider.
What should I ask about the data boundary?
Ask where encryption happens, who holds the keys, which components can see readable inputs, and who can decrypt the result. Assess the specific application and deployment before using confidential information. Start demonstrations with public or invented examples.
Is the network already decentralised?
Independent network participation is in development. The public preview and an available Chat demonstration do not establish an already decentralised network. Operator participation requires technical review, reproduction and published rules.
What do the performance figures establish?
The figures in the evidence section are Aura-reported internal results for specific workloads. They do not establish universal production readiness or independently validated security. Numeric correctness, model performance and key custody require separate evidence.
A more private
future starts
with us.
Want to see FHE Chat or discuss an application idea? Tell Gen what you would like to explore. Include your preferred time and timezone for a live demonstration.
gen@afhe.io ↗