Straight answers. No filler.

Start here
You are not switching, and that distinction matters more than it sounds like it should. DreamerOS runs on top of the AI you already pay for. It does not replace ChatGPT or Claude, it watches their work. Here is the real-world problem it solves: you ask a model a hard question, it gives you a confident, well-formatted answer, and you have no way to know if part of it is quietly wrong, because the model sounds exactly as confident when it is wrong as when it is right. DreamerOS scans every answer for that kind of drift before it reaches you, so a fabricated statistic or a skipped constraint gets caught instead of acted on. The second problem: ChatGPT has no idea what you decided in Claude yesterday, and Claude has no idea what you decided in ChatGPT last week. You end up re-explaining your situation every single time you switch tools. DreamerOS holds one memory that every engine reads from and writes to, so the context follows you instead of staying trapped in one app. Keep your subscriptions. Add the layer that makes them reliable and stops making you repeat yourself.

It means there is no such thing as a wrong or badly-worded prompt, and no cap on how many you can send. Most people who struggle with AI tools are not bad at using them, they are just typing the way humans naturally think: half a sentence, a typo, a thought that trails off. Every "prompt engineering" guide exists to fix that gap by teaching you to write more like a machine wants. DreamerOS closes the gap from the other direction. You talk normally. A component called the Endless Dream Engine takes your messy input and restructures it into an optimized, engine-specific request behind the scenes, before the model ever sees it. One half-formed thought in, an expert-level structured answer back.

Light is 100% free
DreamerOS Light is free, permanently, not a trial. No credit card at signup, no countdown, no artificial message cap designed to nudge you into paying. Use it through the GPT Store with zero setup, or at app.dreameros.app with a free account that takes about thirty seconds. Light includes promptless interaction, automatic thinking-mode detection, follow-up forecasting, and layered answers. The honest reason it can be free: Light runs one engine with warn-mode checks instead of five engines with fail-closed checks, so it costs DreamerOS less to run per user. You are not missing a hidden catch, you are getting a smaller slice of the same architecture.

Two doors into the same house. The GPT Store version is the fastest: search for the DreamerOS GPT, open it, and start typing, no signup screen in the way. The web app at app.dreameros.app asks for a free account first, and in exchange gives you conversation history you can come back to, the intent verification metadata behind each answer, and a clear path to Pro when you want it. Both run the exact same intent-verified pipeline underneath, so the answer quality does not change based on which door you walked through, only how much you can see about what happened.

Only if you want to use the GPT Store entry point. The web app at app.dreameros.app is fully independent: no ChatGPT account, no Claude account, no other AI subscription required at all. And if you already have Claude or another MCP-compatible client installed, you can add DreamerOS as a connector and keep working inside the tool you already have open, rather than opening a new tab.

Fair concern
Fair worry, and the honest answer is: there is a learning curve for the features, not for the basics. Day one, you type the way you already type into ChatGPT, and DreamerOS just quietly checks the work. That part has zero curve, because it is designed to disappear. The curve shows up later and only if you want it to: things like picking which engine handles a question, reading the response metadata panel, or setting up cross-engine memory take a few minutes to understand the first time you look at them. Nobody is required to touch any of that. Think of it the way a car has cruise control you can ignore forever, or learn in one afternoon and never think about again.

This is exactly the situation the system is built for, not an edge case. Most people do not arrive at an AI tool with a perfectly formed question, they arrive with a rough sense of what they need. When your message is ambiguous, DreamerOS does one of two things, and which one depends on the stakes. If a wrong guess would not cost you much, it makes a reasonable assumption, answers, and tells you plainly what it assumed, so you can correct it in one reply. If getting it wrong would actually matter, it asks you one clarifying question instead of guessing. Either way, you are never left holding an answer built on a silent guess you never got to see.

Think of raw ChatGPT or Claude as a very knowledgeable person you just met at a party. They know a lot, they will confidently answer almost anything, and they have absolutely no memory of you from one conversation to the next. DreamerOS adds three things on top of that same person: someone who translates your question into something they can actually use well, someone who decides which of the five specialists in the room is best suited to answer, and someone who fact-checks the answer before it reaches you. The model itself has not changed. What changed is everything wrapped around it.

No, and the distinction is worth being precise about because "wrapper" gets thrown at every AI product these days. A wrapper changes what you see: new buttons, a nicer chat window, the same model doing the same thing underneath. DreamerOS changes what happens to your message before and after the model sees it. Every request goes through computational prompt shaping, gets routed across five engines based on the type of task, and comes back through integrity validation. When that validation layer spots ambiguity or high stakes, an independent engine silently cross-checks the answer without you asking for it. Your input is transformed before the AI sees it. The output is checked before you see it. A new coat of paint does neither of those things.

Core architecture
Three layers, every single message, no exceptions. Layer 1 restructures your raw input into a sharp, complete prompt. Layer 2 is the engine actually generating the response. Layer 3 reads that response back and checks it for hallucination, drift, and gaps before you ever see it. Picture it like a kitchen: layer 1 is the person taking your order and writing it down clearly, layer 2 is the cook, layer 3 is the person who tastes the plate before it leaves the pass. Pro users can see the full metadata from all three layers on every response, so nothing about the process is hidden from you, it is just invisible by default.

It takes what you actually typed and rebuilds it into a structured, domain-aware, engine-specific prompt before the model ever reads it. Concretely, it detects what you are trying to accomplish, fills in context you left out but clearly meant, picks the right depth of thinking for the question, and formats the whole thing for whichever engine is about to handle it. You never see this step happen. But it is the entire reason a seven-word question can come back as a properly structured strategy document instead of a shallow paragraph: the model was never given your seven words, it was given the fully expanded version of what those seven words meant.

Every response gets scanned for five things: hallucination risk, confidence gaps, assumption drift, completeness, and whether the facts might already be stale by the time you read them. The behavior differs by tier on purpose. In Light, a failed check surfaces as a visible warning, so you know to double-check that part yourself. In Pro, it is fail-closed: if a response does not pass, it gets intercepted and reworked before it ever reaches your screen, rather than shipping with a warning label attached to a wrong answer.

When the integrity layer flags a response as ambiguous, low-confidence, or high-stakes, a second, independent engine quietly reviews the answer in the background, the same way a hospital gets a second radiologist to look at an unclear scan. If that second engine finds a real problem, you see a warning. If it agrees with the first answer, you never know the check even ran. This does not happen on every message, only when the system judges the stakes are high enough to justify the extra check, and it runs asynchronously so it adds no wait time to your response.

Four rules that are enforced in code, not written in a style guide someone can ignore under deadline pressure.
Law 001: Structural Precision - your prompt gets sharpened until there is zero ambiguity left in it
Law 001A: Expert Framing - the response is framed the way a domain expert would actually answer, not a generic assistant
Law 002: Complete Delivery - you are never told to "go check elsewhere" for the actual answer
Law 003: TLDR First - the exact answer sits on line one, the reasoning and detail follow after
Law 004: Live Data Rule - if the true answer could be different tomorrow than it is today, the request routes to the Research engine and comes back with real, current sources instead of a guess frozen at training time
These fire on every response. There is no toggle to turn them off.

A set of rules enforced on every single response, not suggestions a model might follow if it feels like it. A few worth naming plainly: accuracy is prioritized over sounding complete, there is no speculative filler added just to pad an answer, no false guarantees get made about outcomes that are not guaranteed, uncertainty gets stated out loud instead of hidden behind confident phrasing, a flawed premise in your question gets pushed back on instead of politely answered anyway, and anything the model would normally quietly drop from a long answer gets flagged instead of vanishing.

The rule that stops an AI from silently deciding what you meant and running with it. When your request is genuinely ambiguous in a way that matters, DreamerOS asks you to clarify instead of guessing. When it has to make a small assumption just to move forward, it tells you exactly what it assumed, right there in the answer, before you act on anything. The goal is simple: you should never be building on top of a guess you were never shown.

Honest answer
Correct, and we would rather say that plainly than pretend otherwise. No verification layer catches everything, and anyone who tells you their AI system is infallible is selling you something. What DreamerOS is actually built to do is reduce the odds of a wrong answer reaching you unflagged, not eliminate them to zero. Every checked answer carries a visible confidence signal, so a lower-confidence response looks different from a high-confidence one instead of both sounding equally sure of themselves. On Pro, every check is logged to an audit trail you can inspect, so if something does slip through, you can see exactly which check ran, what it found, and why it did not catch the problem. That transparency is the actual promise. Perfect accuracy is not something any AI system can honestly promise you.

Think of it like a newsroom instead of one reporter trying to cover everything. A real newsroom does not send its investigative reporter to write the weather update, and it does not send the copy editor out to break a live story. Each person is genuinely better at one kind of work, so the editor routes the story to the right desk. DreamerOS runs five specialized engines the same way, each with a defined job:DreamerOS deliberately never shows you which underlying vendor or model is behind a given role. You interact with the job being done, not the brand doing it, the same way you do not need to know which specific journalist wrote a story to trust the newsroom's editorial process.

By design, not secrecy
Two real reasons, both practical. First, the underlying models change. A vendor ships a better version, and DreamerOS evaluates it and swaps it in when it clears intent verification review, usually within two weeks of a major release. If the marketing promised you "powered by Model X," that promise breaks the moment DreamerOS finds a better fit for the job. Naming the role instead of the vendor means the promise never breaks: Research always means citation-grade search, regardless of which engine is doing the searching this month. Second, most people do not actually want to become AI model experts to get a good answer. You want the weather question answered correctly, not a briefing on which language model is best at weather questions this quarter. DreamerOS carries that complexity so you do not have to.

Yes - live data
Yes, and this is actually a good test case for why routing exists at all. A model trained on data through some past date has no way to know what happened after that date, no matter how confidently it might guess. DreamerOS detects when your question needs information that could be different right now, and automatically routes it to the Research engine, which pulls real, current sources instead of extrapolating from old training data. Weather, stock prices, breaking news, whether a product is still in stock, anything with a shelf life gets routed to live search rather than answered from memory.

On Pro, yes, you can teach DreamerOS which engine you want handling a given kind of task, the way you might tell a team "route anything about contracts to Sam." By default, without you setting anything, DreamerOS picks the best engine for each task automatically based on what the question actually needs. Simple questions get answered fast without invoking the full five-engine machinery. Complex or high-stakes questions get the full pipeline, because that is where the extra checking actually earns its cost.

Four different ways a question can move through the constellation, and each one solves a different real-world need.
Best fit: DreamerOS picks the single optimal engine automatically. Use this for almost everything, it is the default for a reason.
Consensus: The same question goes to multiple engines in parallel and the responses are compared for agreement. Use this when being wrong would actually cost you something, like a decision you are about to act on.
Compare: Each engine's response is shown side by side and you decide which one you trust more. Use this when you want to see the reasoning differences yourself, not just the final answer.
Sequential: The output from one engine becomes the input to the next, a real pipeline handoff. Use this for multi-stage work, like research feeding into a draft feeding into a final polish pass.

DreamerOS analyzes your question before it reaches any engine at all, the way a hospital triage nurse assesses a patient before deciding which department sees them. It classifies the type of task you actually need done, research, building, analysis, or creative work, and sends it to the engine built for that job. This happens before the question-shaping step, not after, because a prompt optimized for deep research looks structurally different from one optimized for producing a finished deliverable. Routing has to come first so the shaping step knows which shape to build toward.

Yes - cross-engine
Yes. In Pro, conversations persist across sessions and across every engine, not just the one you happened to be talking to. Ask something in Build today, switch to Intent Verification tomorrow, and both conversations draw from and write to the same memory. Compare that to the usual experience of juggling multiple AI apps, where every new chat starts from zero and you spend the first few minutes re-explaining your situation before you can even get to the question.

Memory that lives at the DreamerOS layer, not trapped inside any single AI platform's account. When you make a decision in one engine, that decision becomes available to every other engine automatically.No copy-pasting between tools. No starting over. One shared context, five specialized engines drawing from it.

Four categories, and it is worth naming all four rather than being vague about it. Conversations, so you can pick up where you left off. Integrity check events, meaning which engine handled each request, what the checks found, and what it cost in tokens. Memory entries, the decisions, facts, and preferences the system extracts from how you actually work. And your account information. Your message content is never logged to any external system outside your own DreamerOS account, it does not get scattered across vendor logging pipelines you have no visibility into.

Yes
Full JSON or CSV export of every piece of memory, every conversation, and every intent verification document, available on every tier including free. This is a deliberate stance, not a technicality: data portability is treated as a right you already have, not a feature you unlock by paying more. If you ever want to leave, you leave with everything you put in.

Your data stays accessible for export for 90 days after cancellation, plenty of time to grab a full copy if you need it. After that window it is deleted per the retention policy, not kept indefinitely just because it once existed. You can also downgrade to Light at any time rather than canceling outright: Light keeps working normally, but Pro-specific features like cross-engine memory, full fail-closed integrity checks, engine routing, and file absorption become inaccessible until you upgrade again. Nothing gets deleted on a downgrade, it just gets gated behind the tier that unlocks it.

Yes
DreamerOS does not sell user data, does not run ads inside conversations, does not silently swap in a cheaper model without telling you, and does not hand your data to third parties for their own model training. On Pro, your interaction data does feed a personal improvement loop, meaning DreamerOS gets better at understanding how you specifically work, but that loop stays scoped to your account. It makes the product better for you, it does not become someone else's training set.

One private database, scoped to your account, in a single US region, managed directly by DreamerOS rather than fragmented across a dozen vendor systems you would have no way to audit. No other user can reach it, and there is no cross-user leakage between accounts. For anyone who wants the systems answer: single authoritative store, append-only receipts, identity kept in a separate plane from application data. Multi-region redundancy is on the roadmap but not shipped yet, and that is stated here rather than implied. One database, your account, your data.

Full control, both directions
Yes, at whatever granularity you actually want. You can delete a single memory entry, everything tagged a certain way, everything of a certain type, or wipe the entire history and start clean. This is not buried in a support ticket queue either, it is a first-class tool the system itself uses (internally called "forget"), the same way "remember" is a first-class tool for writing new memory. If a decision changes, an old preference no longer applies, or you simply want a fresh start, you are never stuck living with context you no longer want followed around.

Light (Free) - Chat with one engine. 10 free image generations a day. Session persistence - pause a conversation and resume it where you left off. Full audit trail. Memory that carries across sessions starts at Pro.

DreamWeaver Solo ($14.00/mo) - Take one engine seriously. Pick one of: Build (ChatGPT), Review (Claude), Synthesis (Gemini), Signal (Grok), Research (Perplexity). Pre-check and post-check on every answer. Cross-session memory. Full audit trail. 2,048-token responses. 50 checked requests per hour.

DreamWeaver Duo ($24.99/mo) - Two engines that share what they know. Pick any two from the full five-engine lineup. They share memory - what one knows, the other knows. Everything in Solo plus the second engine and the shared layer.

Pro ($39.99/mo) - All five engines, verified. The system picks the best engine per question. Memory compounds across every engine and every session. File / image / audio absorption. Image generation through all five verified engines. MCP connector. API key. Direction Set Workshop. 5,000-token responses. 100 checked requests per hour. Full audit trail with sources, costs, and integrity scores.

Elite (Invite only) - Everything in Pro plus operator-grade verification: custom routing rules, Custom GPT and Custom Claude Project provisioning. The multi-engine image quorum and iterative vision critique loop are coming as wired for Elite.

DreamWeaver is the on-ramp between free and full Pro, built for the person who has one AI tool they actually live in and just wants it checked. Solo lets you pick one engine and get pre-check and post-check on every single answer from it. Duo lets you pick two, and the two share memory with each other. Both run the exact same integrity pipeline Pro uses, just scoped to the engine or engines you chose instead of all five. If you eventually want the full five-engine lineup plus file absorption, image generation, and the Direction Set Workshop, that is when Pro makes sense.

This is the key question
Light gives you one engine with warn-mode checks, meaning it flags a problem and lets you decide what to do about it. Pro gives you all five engines with fail-closed checks, meaning bad output gets intercepted before it ever reaches your screen instead of arriving with a warning label attached. Pro also adds file and image absorption so you can drop in an actual document and reason on it, image generation verified across all five engines, the Direction Set Workshop for building portable operating instructions that work across 66 different AI tools, memory that compounds across engines instead of resetting, and the MCP connector for plugging DreamerOS into tools you already use. The honest pitch: if you use AI for work where a wrong answer would actually cost you something, Pro tends to pay for itself the first time it catches a hallucinated fact you were about to act on.

Because what you are paying for and what DreamerOS does are two different jobs. ChatGPT Plus and Claude Pro are paying for a faster, more capable engine. DreamerOS is the verification layer sitting on top of whichever engine you use. ChatGPT Plus does not check its own output for you, and it has no mechanism to. Claude Pro has no idea what you decided in ChatGPT last week, because the two products were never designed to talk to each other. Neither one routes your specific question to whichever of five engines is actually best suited to answer it. There is also a structural reason no single AI vendor can fully solve this on their own: asking a model to audit itself is asking the fox to watch the henhouse, it has no outside perspective on its own blind spots. DreamerOS does not ask you to give up your subscriptions. It makes the ones you already have more reliable.

Honest answer
If your questions are genuinely low-stakes, like a recipe substitution or a quick grammar check, Light alone covers you fine and you may never need to pay a cent. The math changes once a wrong answer would actually cost you time, money, or credibility: drafting a contract clause, sizing a budget, diagnosing a bug in production code, deciding whether a claim in a report is true before you repeat it in a meeting. That is the line worth thinking about, not "do I use AI a lot" but "what happens the one time it is confidently wrong and I do not catch it." If that answer is "not much," stay on Light. If that answer is "actually, that would be a real problem," that is what the paid tiers exist to catch.

Pro users see the exact token count and USD cost on every single response, right in the intent verification metadata sidebar, so nothing about the cost is hidden after the fact. Typical intent-verified queries run $0.002 to $0.01 depending on complexity and which engine handles them. Your subscription already covers all of the intent verification processing itself, the per-query number shown is the raw underlying engine cost, displayed for transparency rather than billed separately on top of what you already pay.

Start here
The DreamerOS GPT inside the ChatGPT Store. No settings files, no MCP configuration, no pasting a server URL anywhere. Click the link, send a message, sign in when it asks. If the phrase "MCP server URL" made your eyes glaze over reading the connectors below, this is the one built for exactly that reaction, and it runs the same intent-verified pipeline as every other connector.

Two options:
Claude.ai - Go to Claude.ai Settings, find MCP Connectors, add a new connector with the DreamerOS MCP URL. Your API key is in your DreamerOS profile at app.dreameros.app/connect.
Claude Desktop - Open Claude Desktop settings, add the DreamerOS MCP server configuration. Same API key.

Once connected, DreamerOS tools (chat, recall, remember, route, etc.) appear as available tools inside Claude. You use Claude normally - DreamerOS intent verification runs invisibly on every interaction.

Easiest option
Use the DreamerOS GPT in the ChatGPT Store. Click the link, send any message, sign in when prompted. Done. All 12 GPT Actions are active. No MCP, no configuration, no pasting URLs. This is the fastest way to start using DreamerOS.

Live now
Perplexity MCP connector is live. Requires Perplexity Pro, Max, or Enterprise. Go to Settings, click Connectors, click "+ Custom connector" (top right), select Remote, name it "DreamerOS", paste your MCP Server URL from app.dreameros.app/connect, set Auth to None and Transport to SSE. Check the box, click Add. DreamerOS tools appear in every thread. Takes 45 seconds.

Gemini supports MCP through the Gemini CLI and the Google AI API. Open your terminal, run gemini, then edit ~/.gemini/settings.json to add DreamerOS as an MCP server with your URL from app.dreameros.app/connect. Restart the CLI and type /mcp to confirm. Note: the Gemini web UI at gemini.google.com does not yet support custom MCP connectors.

Grok supports remote MCP servers through the xAI API. Pass DreamerOS as an MCP tool in your API call using your MCP Server URL from app.dreameros.app/connect. Both the xAI SDK and the OpenAI-compatible endpoint work. Note: the Grok web UI does not yet have user-facing MCP connector settings.

Yes
DreamerOS exposes 20 tools via the Model Context Protocol on the free tier, rising to 41 on Elite. Any MCP-compatible client can connect using your DreamerOS API key. Same intent verification pipeline. Same integrity checks. Same memory. Wherever MCP works, DreamerOS works. For step-by-step setup for every platform, see the connect page.

20 tools on the free tier, rising to 41 on Elite. Free covers the core loop: chat (intent-verified conversation), recall (search memory), conversations (list history), verify (pressure-test decisions), full memory export, coding context, connector health, and session tools. Paid tiers add remember and forget (write and delete memory), context (full session state), route (multi-engine routing), space (manage workspaces), state (manage cognitive state), canon (query intent verification docs), agent (intent-verified agentic tasks), and cross-service observability on Elite.

Two surfaces, one auth. MCP tools at /api/v1/actions/ for memory and routing (chat, recall, remember, route, verify, and so on). REST endpoints at /api/v1/ for the B2B Trust API: Verify, Underwriter, Bonds, Provenance, Audit Export, Cross-Service Feed, Intent Clarify, Fingerprint Export, Caught-This-Week leaderboard. Both authenticate with the same Bearer token (your API key). Full docs on the API page.

Ten B2B integrity and trust primitives shipped on the gateway 2026-05-19. Tier 1 (consumer): Verify (light free, medium $0.02 and the default, deep $0.05, exhaustive $0.25), Fingerprint Export (cryptographically signed portable passport), Caught-This-Week leaderboard (public, pseudonymous), Cross-Service Feed (Sentry, Railway, Vercel, Supabase, GitHub unified), Intent Clarify (intent classification layer). Tier 2 (B2B and Enterprise): Underwriter (Pro+, $0.10/call), Bond Registry (Pro+, /listings public), Provenance (Pro+, C2PA-aligned), Adherence Audit Export (Elite). All usage-billed through event cost tracking so charges land itemized on your Stripe invoice. Full reference at /dreameros-api.php.

Five engines managed via the Model Routing Registry. DreamerOS never exposes vendor or model names to users - you interact with roles (Build, Intent Verification, Synthesis, Signal, Research). Models are updated within 2 weeks of new GA releases and require intent verification review before reaching production.

A fair worry, since "we check everything" can sound like a euphemism for "we make you wait." In practice: the Second Opinion runs asynchronously in the background, so it adds zero user-facing latency, you get your answer at normal speed and the check either confirms it silently or flags it after the fact. Prompt shaping adds roughly 200ms, imperceptible against normal typing-to-response time. The integrity post-check runs in observe mode, also async. Total visible overhead for a checked response is minimal compared to the raw engine response time you would get anyway.

Not currently. Enterprise tier (custom pricing) includes dedicated infrastructure options. Contact us for details.

Every Pro response includes a metadata panel showing: which engine handled your request, which thinking mode was detected, whether integrity pre-check and post-check passed, whether the Second Opinion flagged anything, token count (in and out), and the USD cost of the request. The audit trail, made visible on every response.

Those are infrastructure plumbing: model routers, API proxies, observability layers, the pipes behind the wall. DreamerOS is an intent verification and experience layer that sits above the plumbing. LiteLLM routes by latency and cost, a purely operational decision. DreamerOS routes by task type, restructures every prompt for the target engine, integrity-checks every output, and maintains cross-engine memory. A router picks which pipe your water flows through. DreamerOS is closer to the water treatment plant checking what comes out the other end. Different category entirely.

No, and the difference is who does the work. Prompt engineering is you, manually, learning to craft better inputs, usually through trial and error over months. The question-shaping pipeline here restructures every input automatically, classifying the domain, calibrating the depth, applying engine-specific optimization, scoring quality, all before you ever hit send. You never write a "good prompt." The system writes the optimized version for you, differently tuned for each engine, on every single message, whether you know anything about prompt engineering or not.

The opposite
Better models need better intent verification, not less. Think about it the way you would think about a faster car: a car that goes 200 miles an hour needs stronger brakes than one that goes 30, not weaker ones, because the consequences of something going wrong scale with the speed. As models get more capable and more confidently persuasive, the cost of an unchecked wrong answer goes up, it does not shrink. DreamerOS is built to evolve alongside the models rather than get left behind by them: the Model Transition Protocol updates routing, recalibrates integrity thresholds, and validates direction sets within two weeks of any major model release.

Full enforcement begins August 2026, with penalties up to 35 million EUR for organizations that cannot demonstrate compliance, and by outside estimates roughly 77 percent of organizations currently lack the infrastructure to actually enforce their own AI policies, let alone prove it to a regulator. DreamerOS is building toward compliance export: audit-ready documentation generated on demand from the same integrity events already being logged for every response. Stated plainly: this is on the roadmap, it is not shipped yet, and it will not be described as done here until it is.

Yes
The DreamerOS GPT runs inside chatgpt.com. Search for DreamerOS Infinite Prompts in the GPT Store, click it, and start using intent-verified AI immediately. Light tier intent verification works for everyone - no account required.

Open the DreamerOS GPT inside ChatGPT and sign in with your DreamerOS account. The GPT detects your subscription tier automatically - DreamWeaver Solo, Duo, or Pro. No reinstall, no configuration. Your tier just activates.

No - free tier available
Anyone can use the DreamerOS GPT for free with Light tier intent verification. Subscribing to DreamWeaver Solo ($14.00/mo), Duo ($24.99/mo), or Pro ($39.99/mo) unlocks additional engines, cross-engine memory, and the full intent verification pipeline - all inside the same ChatGPT interface.

This is the important one
It means DreamerOS operates under explicit rules that cannot be silently changed or quietly skipped when they are inconvenient. No guessing, enforced by the Zero-Assumption Protocol. No silent drops, meaning nothing gets quietly omitted from a long answer without a flag. Integrity checks on every single output, not a sample of them. When a check flags real risk, a second independent engine cross-checks the answer through the Second Opinion. The checks are the architecture itself, not a feature bolted on afterward.

The integrity framework that keeps DreamerOS accountable to itself, not just to you. It makes every AI interaction answer three questions before it counts as complete: Who is asking (Authentication), What is allowed (Authorization), and What actually happened (Accounting). The Buoyancy Score badge in the corner of your screen reflects that integrity health in real time, the way a fuel gauge tells you your actual state rather than what you hope it is.

Yes, three ways, and you do not have to take our word for any of them. (1) Use the free tier and compare the output quality directly against raw ChatGPT on the exact same question, side by side. (2) Upgrade to Pro and see the full response metadata on every single response: which engine handled it, what was checked, what was flagged. (3) Check the running integrity stats: 7,000-plus checks performed and counting, a number that only grows because every check is logged, not selectively reported.

DreamerAI. Built by a systems architect who operates across all five engines daily as the Human Conductor, the person who was doing this work by hand before it became software. DreamerOS automates what that founder did manually every day: route tasks to the best engine, validate output quality, maintain cross-engine memory, and enforce quality rules that do not bend under deadline pressure. The product is that workflow, turned into something anyone can use.

Yes. If you believe your API key has been compromised, contact support immediately and we will rotate it. A self-service key rotation feature in the settings dashboard is on the product roadmap.

For everyone
In the app. Every checked answer's receipt lives at Receipts, and the full signed record of every check lives at The record. Any receipt ID can be verified without an account.

For everyone
Every DreamerOS response can be signed and shared. Anyone with the receipt ID (format dro_evt_*) can verify the integrity verdict and integrity signature at /api/v1/receipts/{id} without a DreamerOS account. Proof of multi-engine verification, on demand. See the API docs.

For businesses
POST /api/v1/verify accepts any text and returns a signed integrity verdict (go / hedge / block) plus the full pipeline result. Other companies use DreamerOS to verify their own AI outputs and ship a signed event_id to their customers. Pro tier and above. See the API docs.

Signature feature
Your cognitive fingerprint (the profile of how you think, decide, and prefer to work, built from your own interactions) is exportable as a cryptographically signed JSON token. Import it into ChatGPT custom instructions, Claude system prompts, Cursor rules, or any MCP-aware AI - those tools instantly pick up how you operate. First portable AI identity across vendors. See the API docs.

Pricing
The Verify API supports four depths: light (free, any tier, verdict only), medium ($0.02 / call, any tier, source-grounded, and the depth you get when you do not name one), deep ($0.05 / call, Pro+, full fidelity axes + signed receipt), exhaustive ($0.25 / call, Elite, full integrity flags + audit trail). Pay only for the depth you need rather than a flat subscription. The live list is always at GET /api/v1/verify/depths. See the API docs.

Public
A public, pseudonymous ranking of users whose DreamerOS pipeline caught the most high-severity or critical decisions in the past 7 days. No PII exposed (user IDs are SHA256-hashed). Proof DreamerOS is catching real risks every week. See the API docs.

For high-stakes work
POST /api/v1/underwrite takes a claim or decision, runs three-axis verification (fidelity, consequence, endgame), and returns a cryptographically signed underwriting certificate with a policy_id. Use it as proof your decision was vetted. Pro+ tier, $0.10 per call. See the API docs.

Public API
POST /api/v1/intent/clarify accepts ambiguous text and returns an intent classification (architectural_decision, ship_it, risk_scan, strategy, day_to_day, and others) with a depth hint. Free for Light tier; Pro+ adds prompt restructuring and emotion signal. See the API docs.

Tracking primitive
A signed snapshot of a users fingerprint state at a point in time. Other AI tools can subscribe to a bond to track that users evolving cognitive profile. Bonds are tracking primitives, NOT financial instruments. Pro+ to issue, public to list. See the API docs.

Notary for ideas
POST /api/v1/provenance/claim takes text and returns a cryptographically signed receipt proving you thought it at that timestamp. Anyone can verify the claim at /api/v1/provenance/{claim_id}. Like a notary for ideas. See the API docs.

For Elite users
GET /api/v1/audit/export returns your full integrity trail (every integrity check event in the window) as a JSON-LD bundle with a tamper-evident bundle_sha256. The complete audit record for any window you specify. Elite tier. See the API docs.