We sent a confirmation link to . Click it and you land right back here — your question stays in the field.
No mail after a minute? Check your spam or promotions folder.
What should the models cross-check?
One answer with its differences and checks.
Pick up where you left off.
We sent a link to
Open it and choose a password. The link brings you straight back here to log in.
Open inboxNothing after a minute? Check your spam folder. If you already have an account, the same link lets you set a new password.
Questions you are waiting on. Checked across AI model families, reported only when the evidence moves.
You can log back in anytime. Your bookmarks stay saved to your account.
consens.io is in active development and currently free to use.
Usage limits apply, and some features are available only to selected accounts. Paid plans are coming. There is no subscription to buy yet.
Feedback or a feature you'd like to use? Write to contact@consens.io.
What consens.io remembers about you across chats. You decide what goes in, and you can change or delete it any time.
Short facts and preferences, saved by Agent or by you. They go with your questions while Memory is on.
Nothing saved yet.
Short, reusable preferences for every question.
Paste a longer memory summary or keep your own background note here. It is plain text; Luna may add or patch one passage only when you explicitly submit a Remember or Correct memory action.
It shapes how you are answered, never what is true — where your memory and the evidence disagree, the evidence wins. It is stored on your account, reaches every device, and is deleted with your account. Watch reruns never use it.
Nothing is collected automatically unless you switch on Let Agent update memory. Then Agent decides what is worth keeping from your own messages — never from web pages, files or emails — and shows every change under its answer with Undo. It does not store passwords, keys or account numbers, and saves health, beliefs or similar sensitive details only when you explicitly ask. Switch it off and nothing new is saved; what was saved stays until you delete it here.
The four About you fields are intentionally short. The server applies your plan’s limit (Free, Plus or Pro) to Saved memories, but a long shared note can pull all six models towards the same framing and make the agreement score look higher than it is. Please leave out health details, beliefs, or other people's personal data.
The shared instruction sent to every selected model. It applies to the question; your memory applies to you.
Customize how the models respond. Leave empty to use the default.
How consens answers your next message.
How Agent compares and checks your questions. Saved in this browser; applies from your next message.
How the app and its results look.
Switch between light and dark mode.
Use the included allowance or connect your own OpenRouter API key.
Manage and test your OpenRouter credential.
Manage permanent account actions.
Permanently delete your account, bookmarks, and stored data. This cannot be undone.
What is consens.io?
consens.io runs your query in parallel across multiple large language models and presents their answers
side by side. In addition, it can integrate external search systems to retrieve current and verifiable
information.
Benefits:
✓ Systematic comparison of Models
✓ Structured consensus building
✓ Integration of web-based research sources
Auto Consensus automatically analyses several model responses, identifies agreements and discrepancies, and generates a consolidated answer. The goal is a transparent and robust synthesis of the available outputs.
The Consensus Button combines the selected model responses into a single joint answer. Before running it, you can include or exclude individual responses so that only the models you consider relevant contribute to the final result.
The Consensus Model is a dedicated model that does not answer user queries directly. Instead, it evaluates, structures, and synthesises the already generated model responses into a coherent overall statement.
Differences highlights substantive deviations between the model responses. This makes it easier to see where models arrive at different conclusions or add distinct perspectives.
With Reasoning on, the same models you selected think longer before they answer. That can help with harder questions, but answers take longer and a run uses more of your daily allowance. It is available on every plan; in Agent mode the same control lets you pick the reasoning level of the chat model.
Web Search forwards your query to a grounding Model in the background. The crawled webpages are passed to all the models. The results are displayed directly within consens.io.
By including or excluding individual model answers, you control which outputs are used in the consensus generation. This allows you to filter out low-quality or irrelevant responses before they influence the final result.
API keys are only required if you want to use your own model access. You need a verified login before saving, testing, or using personal credentials.
consens.io provides access to current large language models from several providers, including OpenAI, Mistral, Anthropic, Grok, DeepSeek, Google Gemini, Moonshot AI/Kimi, Z.ai/GLM and Meta Muse.
Your remaining allowance is shown in the sidebar. Compare, Consensus, Reasoning runs and Agent share it. Usage is limited during early access, and limits may change as the service develops.
consens.io is currently free during early access. Usage limits apply, and some features are available only to selected accounts. Paid plans are coming. There is no subscription to buy yet. For more runs, you can connect your own OpenRouter API key in Settings and pay OpenRouter directly. Feedback is welcome at contact@consens.io.
When the models contradict each other on a checkable fact, a separate judge reads the sources they cited for each position. Under the answer, the Contradictions link says how many of them the sources settled; open it to see each verdict with the original passages and links. Checks run with limits; unavailable evidence and omitted checks are shown explicitly. This is not a complete fact-check of the answer, and model agreement is separate from source evidence. The check is on by default and can be switched off in Settings → Runs; switching it off keeps the answer, its marks and the contradictions.