How Amnesia AI® compares: frontier chat, enterprise AI, running it yourself.

Amnesia AI®Open-weight models, sealed hardwareFrontier AI chatThe strongest models, closedEnterprise AI, contractedFrontier models, zero-retention tierRun it yourselfOn-prem, or a model on your laptop

Cost to start$0, then $19/mo$0, then a monthly seatAn annual contract, per seat, plus the rolloutFree on a laptop you own; GPU capex at any real scale

Model powerOpen-weight, up to 122BFrontierFrontierWhatever your own hardware serves

Ease of setupDownload and use it, nothing to deploySign upProcurement, a contract and a rolloutGPUs, an inference stack and an ops team

PrivacyOpened only inside sealed hardwareReadable, under terms of serviceReadable, under a promise not to train on itNothing leaves your own network

ProofA signed report, every replyNoneAn audit of the vendor, not your answerYou are the auditor

Wins & tradeoffA signed receipt on every attested replyAgentic tools, every outward action approved per callNo offline mode, and the models are open-weight rather than frontier.The strongest models availableNothing at all to set upYour work is readable, and terms of service are the only thing saying it will not be read.Frontier models under a negotiated contractA named counterparty who is liableProcurement and a rollout before the first answer, and the audit covers the vendor rather than your answer.Nothing leaves your own networkWorks with no network at allYou buy the GPUs, run the stack and audit yourself, and a laptop-sized model stays a laptop-sized model.

The dots rate each category against the other three, not against an absolute, and the dollar signs are what it takes to get started rather than a price we would quote for anyone but ourselves — Amnesia AI® figures come from the pricing page. Every rating restates a row of the long table below, where two of the rows are ones Amnesia AI® loses.

▸The long version: nine rows, side by side

Amnesia AI®Open-weight models, sealed hardwareConsumer AI chatFrontier models, closedEnterprise AI, zero-retention tierFrontier models, under contractSelf-hosted / on-premYour GPUs, your ops teamA local model on your laptopWhatever your hardware fits

Privacy & proof

Can anyone read your work?No. Opened only inside sealed hardwareYesYes, with a promise not to train on itNo, if you run and secure itNo

What backs that up?Hardware proof anyone can checkTerms of serviceA negotiated contractYour own hardware and staffYour own machine

Proof with every answer?A signed report of seven checks, exportableNoneAn audit of the vendor, not your answerNo. You are the auditorNot applicable

Leak surface publishedYes, itemized on this pageNoNoWhatever you documentNot applicable

What it takes to get there

What it takes to startLowDownload and use it, nothing to deployLowSign upHighProcurement, a contract and a rolloutHighGPUs, an inference stack and an ops teamMediumInstall a runtime, fetch the weights, tune it

What it costs to start$0, then $19/mo$0, then a monthly seatAn annual contract negotiated per seat, plus deploymentGPU capex or reserved cloud, plus the people to run itFree, if your hardware can already run it

What it can do

Model classOpen-weightHosted open-weight models, up to 122BFrontierFrontierFrontierFrontierOpen-weightAs large as your cluster can serveSmallWhatever your laptop fits

Agentic toolsYesEvery outward action approved per callVariesVaries by productVariesVaries by productBuild itThe stack serves tokens; the agent is yours to writeRareRare

Works with no networkNoNoNoYes, inside your own networkYes

Here the self-hosted and laptop categories are kept apart, which the table above merges. No dollar figure is quoted for any provider but our own: everything else is described by its billing model rather than a number we cannot source.

Compared to a specific tool

vs ChatGPT vs Claude vs Gemini vs Grok vs Ollama

Private AI without the on-prem bill.

Decide on your own files.