The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Confer is Moxie Marlinspike’s take on chatbots that prioritizes privacy above all else: it encrypts conversation history on the user’s device and is designed to process prompts inside an attested Trusted Execution Environment, rather than exposing readable chats to the service operator. Confer is hosted, not fully local, so its privacy depends on client, cryptographic, hardware, and operational trust.
Marlinspike, the creator of Signal, built Confer around a problem that conventional chatbot privacy settings do not completely solve. AI conversations often contain unfinished thoughts, personal insecurities, medical and financial details, work documents, correspondence, and the reasoning behind a decision—not merely a finished question.
Confer’s answer is not to run the entire model on your laptop. The service sends encrypted prompts to remote infrastructure, where verified code inside a confidential-computing environment is intended to process them without exposing plaintext to the infrastructure operator. That makes Confer meaningfully different from ordinary hosted AI, but it does not make Confer anonymous, invulnerable, offline, or automatically suitable for regulated information.
Key takeaways
- Confer is a hosted general-purpose AI assistant created by Moxie Marlinspike, the creator of Signal.
- Confer says conversation history is encrypted before leaving the user’s device, while private inference uses a Trusted Execution Environment and remote attestation.
- Confer’s design is not fully local: remote servers still perform the model inference inside protected hardware.
- Confer advertises image understanding, web search with sources, tools, MCP, user-controlled memory, ChatGPT import, and encrypted folders.
- TechCrunch and TIME reported in January 2026 that Confer’s paid plan cost approximately $35 per month, but that price and the reported free-tier limits require checking before publication.
- Confer is most compelling for people who value provider-blind conversation storage and inference more than maximum model choice, mature enterprise controls, or offline operation.
What is Confer, and who created it?
Confer is a general-purpose hosted AI assistant created by Moxie Marlinspike, the creator of Signal. Confer is designed to apply privacy expectations associated with private messaging to AI chat, while recognizing that an AI model must normally receive a prompt somewhere in order to calculate a response. Confer describes the product and its privacy architecture on its official homepage.
Popular on Amazon
- Please note, this device does not support E-SIM; This 4G model is compatible with all GSM networks worldwide outside of the U.S. In the US, ONLY compatible with T-Mobile and their MVNO's (Metro and Standup). It will NOT work with other CDMA carriers, and it is also not compatible with their MVNO (Visible, Xfinity Mobile, US Mobile, Cricket Wireless, etc).
- Compatibility with certain third-party devices and accessibility accessories, including some hearing aids, may vary depending on manufacturer support, Bluetooth protocols, software compatibility, and regional firmware limitations. For additional hearing aid compatibility information, please refer to Samsung’s official support documentation.
- Camera: 50 MP, f/1.8, (wide), 1/2.76", 0.64µm, AF | 50 MP, f/1.8, (wide), 1/2.76", 0.64µm, AF | 2 MP, f/2.4, (macro). Battery: 5000 mAh, non-removable | A power adapter is NOT included.
The phrase “Signal for AI” can be useful shorthand, but Confer is not simply Signal with a chatbot added. In Signal, both ends of a conversation are users’ devices. In Confer, the model must run on a server. Confer therefore combines encrypted client-side storage and transport with confidential computing intended to hide plaintext from the operator while inference takes place.
Marlinspike’s motivation is that conversational AI encourages unusually intimate disclosure. In his essay “Confessions to a data lake”, Marlinspike argues that people may reveal incomplete ideas, uncertainties, personal insecurities, medical information, financial information, and private documents to an AI assistant. His “data lake” framing is an argument about the privacy risk of conversational AI, not an independently established measurement of every AI provider’s data practices.
What does Confer offer besides privacy?
Confer’s indexed product material advertises a frontier open model, image understanding, web search with sources, tool use, MCP, user-controlled memory, ChatGPT import, folders, and encrypted folders. Confer also says users can start for free without entering a credit card at signup. These are advertised capabilities and can change, so readers should verify availability in the live product.
Confer’s encrypted-folders announcement says folder data and ordering are stored in encrypted form and sorted on the client. Encrypted folders are useful for organizing private material, but they do not remove the risks of a compromised device, an over-permissioned connector, or a user deliberately sharing decrypted content.
Privacy does not equal feature parity. Before replacing ChatGPT, Claude, Gemini, a local model, or a business AI system, compare the capabilities that matter to your work:
| Decision area | Questions to answer about Confer | Why the answer matters |
|---|---|---|
| Model quality | How does the available open-weight model perform on your actual writing, coding, reasoning, and research tasks? | Privacy protection is not a substitute for an acceptable answer quality. |
| Model choice | Can you select among models, or are you limited to the models Confer makes available? | Major AI platforms may offer broader model ecosystems. |
| Context and files | What context-window, image, document, upload, and export limits apply? | Large legal, research, or engineering workflows may depend on these limits. |
| Search and citations | How reliable are web-search results and their sources for your use case? | “Web search with sources” does not guarantee accurate or complete research. |
| Integrations | Are the required connectors, MCP tools, API features, calendars, drives, or business systems supported? | Connectors add usefulness but also expand the data and prompt-injection threat model. |
| Administration | Are team controls, audit reports, contractual terms, retention details, and compliance documentation available? | Professional and regulated use often requires evidence beyond consumer product claims. |
| Access | Does Confer meet your mobile, desktop, browser, cross-device, and recovery requirements? | Encrypted key management can make account and device recovery more consequential. |
How does Confer’s privacy model work?
Confer’s privacy design has several separate layers. Encrypted history, encrypted network transport, and private server-side inference solve different problems; one layer should not be treated as proof that all the others are secure.
1. Conversation history is encrypted on the device
According to Confer’s privacy policy, conversation history is encrypted before leaving the user’s device, and the encryption keys are generated and stored on the user’s device. Confer may store encrypted history on its servers, but the company says it cannot decrypt the stored conversations.
That claim primarily addresses data at rest and the operator’s access to saved chats. It does not mean that a browser, phone, operating system, malicious extension, spyware, or another person with access to the unlocked device cannot see plaintext while the user is reading or composing a conversation.
2. Passkeys help derive encryption key material
Confer’s technical explanation of passkey-based encryption says the design uses WebAuthn passkeys and the WebAuthn PRF extension to derive durable secret material. The stated goal is to make cross-device encrypted access easier without requiring every user to write down and protect a separate recovery seed phrase.
The technical post describes a 32-byte PRF output from which client-side encryption keys can be derived. Support varies by operating system, browser, password manager, and authenticator. The post describes stronger support on recent macOS, iOS/iPadOS, Android, Chrome, Safari, and Firefox combinations, while Windows and Linux may require a compatible password manager or authenticator. Because platform support changes, users should check current Confer documentation rather than treating the December 2025 compatibility information as a permanent guarantee.
Passkey-based key management also changes the user’s responsibility. If every authorized device and the relevant passkey are lost or deleted, encrypted history may be unrecoverable unless Confer’s current recovery process provides another secure route. Confer’s passkey design is intended to improve access across devices, but the available research does not establish what happens in every lost-device, deleted-passkey, or account-recovery scenario.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
3. Private inference uses a TEE
Confer’s private-inference explanation says encrypted prompts and responses travel to a confidential virtual machine or Trusted Execution Environment, where approved code performs inference. The host operating the machine is intended not to see the TEE’s memory or execution state.
Also popular on Amazon
- YOUR CONTENT, SUPER SMOOTH: The ultra-clear 6.7" FHD+ Super AMOLED display of Galaxy A17 5G helps bring your content to life, whether you're scrolling through recipes or video chatting with loved ones.¹
- LIVE FAST. CHARGE FASTER: Focus more on the moment and less on your battery percentage with Galaxy A17 5G. Super Fast Charging powers up your battery so you can get back to life sooner.²
- MEMORIES MADE PICTURE PERFECT: Capture every angle in stunning clarity, from wide family photos to close-ups of friends, with the triple-lens camera on Galaxy A17 5G.
- NEED MORE STORAGE? WE HAVE YOU COVERED: With an improved 2TB of expandable storage, Galaxy A17 5G makes it easy to keep cherished photos, videos and important files readily accessible whenever you need them.³
- BUILT TO LAST: With an improved IP54 rating, Galaxy A17 5G is even more durable than before.⁴ It’s built to resist splashes and dust and comes with a stronger yet slimmer Gorilla Glass Victus front and Glass Fiber Reinforced Polymer back.
According to Confer’s technical explanation, the architecture uses confidential computing, a TEE, Noise-based handshakes, remote attestation, reproducible builds involving Nix and mkosi, a transparency log containing signed releases, ephemeral session keys, and forward secrecy. These mechanisms are intended to make the server-side processing confidential and make silent substitution of arbitrary inference code more difficult.
4. Remote attestation checks the protected endpoint
Remote attestation is the mechanism intended to let the client verify that it is communicating with an approved protected environment. Confer says the client checks that the endpoint uses genuine TEE hardware, that the attested public key matches the key used for the encrypted handshake, that measurements correspond to an approved release, and that the release appears in a publicly auditable transparency log.
Attestation verifies a particular hardware and software state; attestation does not prove that the software is bug-free, that the model is unbiased, that the browser client is uncompromised, or that every connected tool behaves safely. A reproducible build can improve confidence that a published source corresponds to a deployed binary, but reproducibility does not by itself eliminate vulnerabilities or prove that the surrounding deployment is correctly operated.
Recommended Free Tools
Is Confer really end-to-end encrypted if the model runs remotely?
Confer can use “end-to-end encrypted AI” to describe a system in which the client encrypts data and verified code inside the remote inference environment decrypts and processes the data, but the model still runs on remote servers. Confer is therefore not equivalent to fully local inference on a user’s own computer.
In a conventional hosted chatbot, the provider’s service receives the prompt in a form that the provider’s infrastructure can generally process, retain, inspect, or route according to its policies and technical controls. Confer’s stated design inserts a protected computation boundary: the prompt is intended to remain encrypted until it reaches approved code inside the TEE, and the infrastructure operator is intended to remain unable to read the plaintext.
The privacy claim consequently depends on more than the encryption algorithm. It depends on the client encrypting the correct data, passkey-derived keys remaining protected, the handshake authenticating the correct endpoint, the TEE hardware enforcing its isolation, the attestation verifier checking the correct measurements, the transparency log accurately representing approved releases, and Confer operating the system as documented.
| Privacy layer | What Confer says the layer does | What the layer does not prove |
|---|---|---|
| At rest | Encrypts conversation history before it leaves the user’s device. | It does not protect plaintext displayed on a compromised device. |
| In transit | Uses encrypted channels, including a Noise-based handshake and forward secrecy according to Confer’s technical explanation. | It does not make an endpoint or connector trustworthy. |
| During inference | Uses a TEE or confidential VM intended to hide prompts and responses from the host operator. | It does not prove that the model or protected code is error-free. |
| Attestation | Checks hardware identity, key binding, release measurements, and transparency-log inclusion. | It does not prove that the client, model behavior, or integrations are safe. |
| Account and keys | Uses passkey-related material for durable client-side encryption keys. | It does not remove the consequences of a stolen account, lost passkey, or insecure recovery process. |
What does Confer protect against, and what does it not protect against?
Confer’s stated architecture is primarily a provider-confidentiality design: it is intended to prevent Confer’s ordinary infrastructure operator from reading conversation content. Overall security is broader than provider confidentiality.
| Threat | What Confer’s design is intended to address | Remaining limitation |
|---|---|---|
| Provider reading saved chats | Client-side encryption is intended to leave the provider with ciphertext. | Plaintext can still exist on the user’s device and in the user’s view. |
| Provider reading prompts during inference | TEE-based private inference is intended to hide plaintext from the host operator. | The claim depends on hardware, attestation, client, build, and deployment security. |
| Network interception | Encrypted transport and forward secrecy are intended to protect the connection. | Compromised endpoints can capture data before encryption or after decryption. |
| Malware or browser extensions | No server-side privacy mechanism automatically controls the user’s device. | Spyware, malicious extensions, and compromised operating systems may access plaintext. |
| Stolen account or passkey | Passkeys can strengthen authentication and key derivation. | Account takeover or loss of all authorized keys can expose or permanently block access. |
| Subpoenas | Confer says it cannot provide readable message content if its architecture has only ciphertext. | Account, payment, authentication, connection, and technical metadata may still exist. |
| Connectors and tools | Confer says it cannot read content from connected services such as Google Drive, Calendar, or Gmail. | External services, permissions, documents, and tools introduce separate privacy and security risks. |
| Bad AI output | Encryption can protect content confidentiality. | Encryption does not make answers accurate, safe, unbiased, or professionally suitable. |
Confer’s privacy policy says the service retains technical information needed to establish connections and deliver messages. The same policy says Confer uses Google Sign-In and Stripe, whose own policies govern those interactions. Confer’s terms also disclaim uninterrupted, secure, or error-free service. These qualifications are why “the provider cannot read message content” is more accurate than “Confer stores no user data.”
Confer says it cannot read messages, use conversations to train AI models, sell or rent personal data or content, monetize conversations, or provide readable conversation records to third parties. Those are company claims supported by the privacy policy and product description. Confer says it cannot use conversations for model training because the company says it cannot read the conversations in the first place.
The statement about training should be separated into data categories. Conversation content is different from technical telemetry, account and payment data, feedback a user explicitly submits, data sent to an external tool, and public web-search requests. The current privacy policy and architecture must be consulted to determine how each category is handled.
Can Confer’s web client or connectors defeat the privacy model?
A server-side TEE cannot automatically protect plaintext that a compromised browser client has already exposed. If a provider serves a modified or compromised JavaScript application, the application could theoretically mishandle data before encryption or after decryption; this is a general web-client architectural risk, not evidence that Confer has been compromised.
Web search, MCP, email, calendars, drives, and other connectors also widen the threat model. A malicious document or web page may contain prompt-injection instructions. A connector may have broader permissions than intended. An unrelated personal or corporate file may be included accidentally. Data may also be sent to a third-party service whose privacy and security practices are separate from Confer’s transport and inference protections.
Also popular on Amazon
- YOUR CONTENT, SUPER SMOOTH: The ultra-clear 6.7" FHD+ Super AMOLED display of Galaxy A17 5G helps bring your content to life, whether you're scrolling through recipes or video chatting with loved ones.¹
- LIVE FAST. CHARGE FASTER: Focus more on the moment and less on your battery percentage with Galaxy A17 5G. Super Fast Charging powers up your battery so you can get back to life sooner.²
- MEMORIES MADE PICTURE PERFECT: Capture every angle in stunning clarity, from wide family photos to close-ups of friends, with the triple-lens camera on Galaxy A17 5G.
- NEED MORE STORAGE? WE HAVE YOU COVERED: With an improved 2TB of expandable storage, Galaxy A17 5G makes it easy to keep cherished photos, videos and important files readily accessible whenever you need them.³
- BUILT TO LAST: With an improved IP54 rating, Galaxy A17 5G is even more durable than before.⁴ It’s built to resist splashes and dust and comes with a stronger yet slimmer Gorilla Glass Victus front and Glass Fiber Reinforced Polymer back.
Confer’s privacy material discusses connecting AI assistants to email, documents, and calendars without exposing private data to an ordinary model provider. That claim should be read narrowly: Confer’s layer may protect the connection to its inference system, but the connected service, permissions, tool implementation, and content being retrieved remain relevant. Give connectors the narrowest permissions possible and avoid treating “Confer cannot read the connector data” as “no external system can access the data.”
How much does Confer cost?
Confer’s homepage says users can start for free without a credit card. TechCrunch reported in January 2026 that the free tier was limited to 20 messages per day and five active chats, while the paid plan cost approximately $35 per month and added unlimited access, more advanced models, and personalization. TIME separately reported a $34.99 monthly paid tier in January 2026.
The January 2026 reports do not establish the price on August 16–18, 2026. Confer’s live signup or billing page should be checked immediately before publication for the current geography, currency, taxes, billing interval, message limits, model access, and cancellation terms. The safest current wording is that Confer is free to start and that approximately $34.99–$35 per month was an early reported price, not a confirmed current price.
Free tools Windows power users keep installed
One-click scans. No signup required.
| Plan signal | Evidence available | What remains to verify |
|---|---|---|
| Free access | Confer’s homepage says the service is free to start and requires no credit card at signup. | Current message, chat, model, feature, and rate limits. |
| Free-tier limits | TechCrunch reported 20 messages per day and five active chats in January 2026. | Whether those limits still apply and to which geography or account type. |
| Paid tier | TechCrunch reported approximately $35 monthly; TIME reported $34.99 monthly in January 2026. | Current price, taxes, currency, billing interval, included models, and cancellation policy. |
Confer’s price represents more than additional messages. The value proposition includes a different data-governance model, confidential inference infrastructure, encrypted storage and key management, and reduced reliance on advertising or behavioral monetization. The counterargument is that privacy claims are harder for ordinary users to audit than a price tag, while Confer may offer less model choice or lower capability than larger competitors.
Is Confer open source and independently audited?
Confer describes its system as open source on its homepage, but the available research does not establish the exact repositories, licenses, production-build correspondence, full inference image, deployment configuration, model supply chain, or an independent security audit. A precise evaluation should therefore say that Confer describes its system as open source rather than implying that every component has been independently inspected.
Open-source code can improve reviewability, but a privacy system also includes hardware, build pipelines, release signing, attestation verification, key management, deployment configuration, browser code, dependencies, operational controls, and incident response. A public repository is not by itself proof that the running service matches the repository or that the complete system is secure.
WIRED’s coverage of the Meta relationship highlighted open questions around documentation, threat modeling, and supply-chain transparency. Until Confer publishes or identifies more independently verifiable evidence, users should treat the architecture as a technically serious stated design that still requires trust in implementation and operation.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteHow does Confer compare with ChatGPT, Claude, Gemini, local AI, and Lumo?
Confer’s key distinction is not that it is automatically more capable than every competitor; its distinction is that it is designed to reduce the provider’s ability to read conversation content. Mainstream hosted assistants generally compete through model quality, model selection, integrations, and mature ecosystems, while Confer makes privacy architecture the central product proposition.
| Option | Where inference runs | Main strength | Main trade-off |
|---|---|---|---|
| Confer | Remote infrastructure inside a TEE, according to Confer’s technical explanation. | Provider-blind encrypted history and intended private inference. | Not offline or fully local; model choice, capability, pricing, documentation, and enterprise controls require close evaluation. |
| ChatGPT | Hosted provider infrastructure. | Broad product ecosystem, model reach, and integrations. | Users must evaluate current privacy controls and plan terms; it is not presented here as equivalent to Confer’s operator-blind architecture. |
| Claude | Hosted provider infrastructure. | General-purpose and coding-oriented hosted AI capabilities. | Users must evaluate current provider policies; it does not provide the specific Confer architecture described here. |
| Gemini | Hosted provider infrastructure. | Large-provider ecosystem and integrations. | Privacy controls and data practices depend on current product and plan terms. |
| Ollama | Local machine or self-hosted environment. | Prompts can remain under the operator’s direct control. | Users manage hardware, models, updates, networking, logs, and security themselves. |
| LM Studio | Primarily local desktop inference. | Accessible interface for running local models. | Requires suitable hardware and may provide less cloud-scale capability or convenience. |
| Proton Lumo | Hosted privacy-oriented service. | Natural alternative for users already invested in Proton’s privacy ecosystem. | Current architecture, retention, model selection, and plan terms must be compared from Proton’s documentation; equivalence to Confer’s TEE design should not be assumed. |
Ollama and LM Studio are the clearest alternatives for users who require prompts and responses to stay on their own machine. Local inference avoids sending the prompt to a hosted provider, but local operation shifts responsibility to the user for server security, model downloads, authentication, logs, backups, network exposure, updates, and connected tools.
Proton Lumo is a closer hosted comparison because it also markets privacy as a central feature. The meaningful comparison should cover current retention, encryption, model selection, connector behavior, and subscription terms rather than assuming that all privacy-focused AI products use the same architecture.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What does the Meta announcement mean for private AI?
In March 2026, Marlinspike announced that Confer would remain independent while its privacy technology was being integrated into Meta AI. The announcement indicates strategic interest in confidential AI, but it does not prove that Meta AI already provides Confer-level end-to-end encryption or that every Meta AI product will inherit Confer’s user-facing privacy model.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →The announcement did not specify which Meta products would receive the technology, which models would run inside TEEs, how Meta would handle identity, metadata, advertising, personalization, safety systems, or auditing, or when users would receive the feature. The accurate description is an announced collaboration or integration plan, not a completed deployment with fully documented guarantees.
Also popular on Amazon
- Charger NOT Included, 6.7" Super AMOLED FHD+, 90Hz Refresh Rate, 385 ppi, 800 nits (HBM), 1080x2340px, 5000mAh Battery
- 128GB, 4GB RAM, microSDXC, Exynos 1330 (5nm), Octa-Core, Mali-G68 MP2 or Mali-G57 MC2 GPU
- Rear Camera: 50MP, f/1.8 (wide) + 5MP, f/2.2 (ultrawide) + 2MP, f/2.4 (macro), LED flash, panorama, HDR; Front Camera: 13MP, f/2.0, Android 14, up to 6 major Android upgrades, One UI 6.1
- 3G: HSDPA 850/900/1700(AWS)/1900/2100; 4G LTE: 1/2/3/4/5/7/12/13/14/20/25/26/28/29/30/38/39/40/41/48/66/71, 5G: 2/5/25/41/66/71/77/78 SA/NSA/Sub6/mmWave - Nano-SIM + eSIM
- US Model – Global Connectivity – Compatible with Most GSM Carriers like T-Mobile, AT&T, MetroPCS, etc. Will Also work with CDMA Carriers Such as Verizon, Straight Talk.
The potential significance is substantial: if a major AI platform adopts verifiable confidential inference, privacy-preserving computation could become a mainstream product feature rather than a niche service. The practical significance remains unknown until Meta documents the exact implementation and users can verify the applicable product behavior.
Who should use Confer?
Confer is a strong fit for privacy-conscious users who regularly use AI for sensitive brainstorming and want to reduce the chance that an ordinary provider can read their conversations. Confer may suit journalists, researchers, lawyers, clinicians, and professionals handling personal or workplace context when provider confidentiality matters more than maximum model performance—but those users still need to check their professional, contractual, and regulatory obligations independently.
Confer is also a reasonable option for users who prefer open-weight models, want a hosted experience rather than managing local hardware, and can protect their devices, browsers, passkeys, and accounts. Confer may be worth paying for when the alternative is avoiding useful AI assistance because ordinary hosted chat feels too exposed.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsConfer is a poor fit for users who need the strongest available reasoning or coding model for every task, extensive business integrations, mature team administration, independently documented compliance, a guaranteed stable service, or fully local and offline operation. Confer’s terms disclaim uninterrupted, secure, or error-free service, and the available research does not establish a documented enterprise compliance program or public API and deployment model for every professional use case.
How should you evaluate Confer for your threat model?
The right question is not simply whether Confer is private. Ask what Confer is private from, at which stage, under which assumptions, and with what independently verifiable evidence.
- At rest: Confirm how conversation history is encrypted, where keys are stored, and what happens when every authorized device or passkey is lost.
- In transit: Confirm how the client authenticates the inference endpoint and whether the current implementation uses the documented encrypted handshake and forward secrecy.
- During inference: Determine whether the TEE, attestation process, release measurements, and transparency log provide enough assurance for the sensitivity of your data.
- At the client: Secure the operating system, browser, extensions, phone, computer, clipboard, screenshots, downloads, and backups. A private server cannot protect a compromised endpoint.
- At the account layer: Protect Google Sign-In, passkeys, authenticators, recovery methods, and session access. Confer’s privacy architecture does not prevent someone who controls your account or device from using the service as you.
- At the integration layer: Review every connector’s permissions and assume that email, calendars, files, web pages, and MCP tools can introduce prompt injection or unintended disclosure.
- At the governance layer: Look for current source repositories, licenses, reproducible-build evidence, independent audits, incident reporting, retention details, enterprise terms, and compliance commitments.
- At the model layer: Test the actual model on representative tasks. Private inference does not guarantee accurate, unbiased, safe, or legally defensible output.
Bottom line: is Confer worth using?
Confer is one of the more technically distinctive attempts to make hosted AI less dependent on trusting the operator with readable conversations. Its combination of device-side encrypted history, passkey-derived key material, encrypted transport, TEE-based inference, remote attestation, reproducible-build claims, and transparency logging is substantially more specific than a simple “we do not train on your chats” setting.
Confer is not a magic privacy shield. Confer is not fully local, offline, anonymous, immune to subpoenas, or protected from malware, account takeover, malicious connectors, prompt injection, implementation bugs, or model errors. The service says a subpoena would produce ciphertext rather than readable conversation content, but that claim concerns message content and does not necessarily cover account, payment, authentication, connection, or other technical metadata.
For privacy-first personal use, sensitive brainstorming, and users willing to accept a newer hosted product and potentially higher cost, Confer is a credible option to test. For maximum capability, formal compliance, broad enterprise administration, or offline operation, a major hosted platform, a locally run model, or a self-hosted deployment may be a better fit. The decision turns on the threat model: private from whom, during which stage, and with what evidence?
Frequently Asked Questions
Does Confer run AI models entirely on my device?
No. Confer is a hosted AI service that sends encrypted prompts to remote inference infrastructure. Confer says the prompts are processed inside a Trusted Execution Environment, but Confer is not fully local or offline.
Can Confer provide my readable chats if it receives a subpoena?
Confer says its architecture leaves the service unable to provide readable conversation content because stored conversations are encrypted and inference is designed to remain private. That claim does not mean Confer has no account, payment, authentication, connection, or technical metadata, and it does not make Confer immune to legal requests.
Is Confer suitable for medical, legal, or regulated data?
Confer may reduce provider access to readable conversation content, but the available research does not establish independently verified regulatory compliance, enterprise contractual guarantees, or a complete audit. Professionals should obtain appropriate assurances and follow their organization’s data-handling rules before submitting regulated information.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesWhat happens if I lose my Confer passkey or every authorized device?
Confer’s passkey design is intended to support encrypted access across devices, but the available research does not establish the recovery outcome for every lost-device or deleted-passkey scenario. Users should verify the current recovery process before storing information they cannot afford to lose.
Is local AI more private than Confer?
Local AI can keep prompts off hosted provider infrastructure, which is a stronger fit for users who require local or offline processing. Local AI is not automatically secure: the operator must protect the computer or server, models, logs, backups, authentication, network, updates, and connected tools.
