Everything Vanyx does.

Vera doesn't just talk — she runs the work, chaining Python simulations, formal proofs, literature searches, and symbolic math into a single grounded answer. Every tool run shows up as a rich card right inside the conversation, so you can watch the evidence assemble in real time. Nothing is a black box, and results are saved so you can build on them later.

Auto uses the house model and routes tools to the task: Lean for supported formal claims, Z3 for constraint problems, and Python for computations. Tool availability, funding, and your workspace controls still apply. A tool result is evidence of what ran; it does not automatically verify the whole answer or contribute your work to a shared knowledge base.

The chatbar offers exactly the choice that matters: Vera 1 (the full-depth default) and Vera Velocity (fast and inexpensive). Both are the same Vera — same tools, same verification, same honesty — and your pick sticks per project. Under the hood, formal goals still route to specialist provers automatically; collapsing the menu is presentation, never capability.

Switch Vera's abilities on or off for each chat: run Python, build interactive surfaces, do symbolic math, write formal proofs in Lean, solve logic with Z3, search the literature, query math databases, remember verified facts, and design quantum circuits. Mix and match exactly what a problem needs, and your choices are remembered for next time. Advanced abilities like quantum stay tucked away until you want them.

Attach up to three documents per message — PDFs, Word documents, spreadsheets, or plain text up to 4 MiB each — by dragging, pasting, or browsing. Vera pulls out the actual text and data so she can reason over your document directly. Documents with too much extracted text need to be split into sections.

Drop in a PNG, JPG, or other image and Vera can read the graph, interpret a handwritten note, or reason over a diagram or photo — all inline, with no extra tools. It's ideal for making sense of visual results, whiteboard snapshots, or figures from a paper. She sees exactly what you see.

Say "pull up the blackboard and show me Einstein's field equations" and a glass chalkboard appears with the equations writing themselves in as Vera explains — real typeset math, sketches, arrows, and callouts in six chalk colors. You draw on the same glass with your finger or mouse, open up to ten boards at once (each addressable by voice — "board 3"), and save any board straight into your knowledge library. Everything written on a board flows into the conversation, so Vera always knows what's on the glass.

Ask for your agents and the panel opens; ask for a derivation and a blackboard appears; ask to keep a simulation handy and she docks it beside the chat. A subtle pulse shows when Vera is driving, and a single autopilot toggle puts her in manual mode where she describes actions instead of taking them — you are always sovereign.

Share any window and a crisp snapshot lands in your next message, giving Vera a clear view of what's on your screen. From there she can help debug, explain, or act on whatever you're looking at. No describing it in words — just show her.

As Vera works, every tool run renders as a clean, purpose-built card inside her reply — arXiv results show titles, authors, dates, and PDF links; Semantic Scholar surfaces citation counts; LMFDB and OEIS display their structured math data; and extracted data lands in tidy tables. Math is beautifully typeset and code is syntax-highlighted. One-click copy buttons let you grab any code or data instantly.

Every conversation saves automatically and reopens with all its tool results still in place, so you can pick up right where you left off. Keep things organized by pinning important chats, renaming them inline, archiving ones you're done with, or deleting them outright. Your research history stays at your fingertips.

Lean checks the declaration encoded in your source. Vanyx screens holes, untrusted assumptions, and compiler-trusting shortcuts. A clean chat-tool result remains compiled but unattested: it does not establish that the declaration matches your intended claim. Only the formal Campaign workflow can earn GOLD through its current verifier policy and exact statement witness; retained historical results are labeled separately.

The loop drafts source, reads the exact errors, and revises until it obtains a clean declaration, reaches its budget, or stops making progress. A clean generic loop result remains compiled but unattested, with no verified claim or proof hash: compilation alone does not show that the source formalizes your intended statement. Failed and inconclusive attempts remain visible.

A proof that quietly assumes its own conclusion looks like it compiles but proves nothing — so after a proof passes, we list exactly which foundational assumptions each theorem relies on and flag anything beyond the small trusted core the Lean kernel is built on. Shortcuts that trust the compiler rather than the logic are flagged too and can never earn the top "gold" rating. When the Lean compiler version is observed, a checked proof gets a compact fingerprint identifying its source and compiler version. Reproducing a run requires the matching toolchain and library environment.

The configured sandbox template preloads Mathlib so proofs can build on its mathematical library. The inline chat path requires that template and refuses an unconfigured environment. Compilation time depends on the source and sandbox; no fixed latency is promised.

Finding the right lemma in a library of over 100,000 usually means hours of manual digging. Instead, describe what you want — "sum of the first n natural numbers," a partial name, or the shape of the statement — and Loogle returns the matching Mathlib results with their exact names, statements, and where to import them from. Searching by the shape of a statement is especially powerful for surfacing every theorem that fits a given form.

Before investing days in a full proof, hand a claim to Z3 as a fast sanity check: it either finds actual values that satisfy your constraints, proves the constraints are contradictory, or hunts for a concrete counterexample that falsifies a conjecture. Many wrong conjectures get knocked down in seconds this way, saving you from chasing a proof of something false. It handles integers, real numbers, bit-level arithmetic, and nonlinear problems, and always halts within a time limit so a hard problem can never hang.

Powered by Wolfram Alpha, this returns exact closed-form results for problems like integrating x² sin(x), solving a quartic equation, or evaluating a special function, plus curated scientific data such as physical constants. Ask for steps and you get a worked solution trace, not just the answer. It complements the solver and proof tools: use it to double-check numerical results from simulations and to crack symbolic problems a constraint solver can't reach.

Three concurrent AI referees focus on rigor and methodology, novelty and significance, and reproducibility and evidence. They have solver, Wolfram, arXiv, or Python tools available to attempt checks in their respective areas. Each returns a recommendation (accept, minor, major, or reject) with a confidence level, and an editor combines their critiques into a revision plan. Tool use is not guaranteed: assess computational and formal claims against the recorded checks, not the recommendation alone.

Vera generates competing hypotheses and gathers solver, symbolic, computational, and literature evidence, with bounded Lean attempts where appropriate. Verdicts distinguish supported, refuted, and inconclusive candidates. Generic natural-language tournaments cannot mint GOLD: a compiled declaration without a separately bound statement remains unattested, and a supported candidate is not a machine-verified discovery.

A bounded set of DOI records is checked for structured retraction notices targeting each requested DOI. An exact retraction relationship is shown as a warning. Missing metadata or an unavailable lookup leaves the status unconfirmed; absence of a warning does not establish that a paper is unretracted.

Inspect each attempt's source, errors, open goals, timestamp, and available provenance. The panel distinguishes compiled but unattested declarations and retained historical results from current verification authority. A historical pass or a clean generic chat compile does not certify the intended claim.

A configured PhysLean/PhysLib template can extend the formal mathematics available for physics work. This capability remains in progress and is not switched on by default. A formal theorem establishes only the encoded statement under its declared assumptions; it does not certify that a physical model matches measured reality.

Vera writes and executes Python in an ephemeral E2B sandbox with scientific packages. Interpreter, request, and sandbox-lifetime limits bound each attempt, with cleanup requested afterward. The provider-default environment and network policy are not sealed reproduction evidence. Outputs report what the submitted code did; they do not by themselves prove a scientific claim, and an EXACT label printed by code is not independent attestation.

When your code generates a chart or figure, it shows up right in the chat next to the results, no screenshots or downloads required. Line plots, histograms, 3D surfaces, and interactive visualizations all render in place as the simulation runs. It feels like watching the analysis unfold in front of you.

Describe or write a quantum circuit and run it directly in the conversation. Add measurements to get realistic sampled outcomes (up to 8,192 shots), or leave them off to see the exact probabilities. It supports the full range of real quantum work, from superposition and entanglement to classic algorithms like Grover's search and the quantum Fourier transform.

When hardware access and provider credentials are enabled, Vanyx attempts a bounded IBM Quantum run. A simulation fallback is explicitly labeled, including known or uncertain hardware submission state. A simulated result is never represented as a hardware measurement, and provider availability or a timeout can leave the hardware outcome unresolved.

Runs that execute Python can retain code, printed output, and figures in a Jupyter notebook. Project notebooks also expose observed environment, source, output, and dataset hashes when available. The current rerun lane is COMPUTED_UNVERIFIED: its mutable provider environment and unsealed dependencies do not establish reproduction, bit-identical results, or independent confirmation.

Vera proposes comparison fields from the paper text available in your conversation. Tables preserve the supplied values and available source links, but do not independently validate retrieval or cell accuracy. Check each value against the original paper before citing it; information the text does not provide should be marked not reported.

Just describe what you want to see and a self-contained interactive visualization renders instantly in the conversation, no download or setup required. Drag sliders, watch animations, and explore simulations that respond to you in real time. Each surface stays in your chat history and comes right back when you reload the page.

One click (or one ask) pops a surface out into its own window — drag it anywhere on screen and resize it from the corner. From there Vera can drive it herself: pause it, restart it, and adjust the very parameters the simulation exposes, narrating as she goes. It turns a visualization into a live instrument the two of you operate together.

Every surface comes with built-in controls to freeze an animation on any frame, replay it from the start, or pop it out into a floating window you can move and resize. It puts you in the driver's seat of the simulation. The chat card stays put, so you never lose the original.

Surfaces can build full three-dimensional worlds on demand, from spinning molecules and orbital mechanics to interactive geometry you rotate with your mouse. Lighting, cameras, and smooth animation are all handled for you. It runs at full frame rate right inside the chat.

Ask for a visualization and surfaces produce publication-quality charts, scatter plots, histograms, heatmaps, and even 3D plots, all with built-in zoom, pan, and hover-to-inspect. Custom data-driven graphics with live transitions and interactive filtering are just as easy. Everything scales cleanly to fit the surface and respects the playback controls.

The frame is separated from your session and cannot contact arbitrary websites. Supported data from its current Project can reach a surface through Vanyx’s governed read bridge. This access is limited to the named Project reads Vanyx supports.

A surface can include datasets and configuration when it is created. Supported Project reads use the governed data bridge and can show loading or unavailable states.

Each surface can be styled with your chosen accent color and adapts its typography to read well at any size. The transparent background lets it sit naturally in the flow of the conversation. The result feels like a native part of your workspace, not a bolted-on frame.

A surface automatically grows and shrinks to match whatever it's showing, staying neatly framed as content animates in and out. You get a clean, seamless embed every time, with no clipped visuals or dead space. It just fits.

When something goes wrong inside a surface, it displays a visible, easy-to-read error notice right where the surface would be. The rest of your conversation keeps working normally, completely unaffected. Problems stay contained so nothing else breaks.

Search more than 220 million papers by describing what you actually mean, not just matching keywords, so the most relevant work rises to the top even when it uses different terminology. Each result shows citation counts, how influential a paper is, and a link to a free full-text PDF when one exists. Narrow by year to focus on the most recent thinking or trace an idea back to its origins.

Ask Vera to search arXiv's full preprint archive by topic or subject area and get back up to 25 clean, structured papers sorted with the newest first. Every result includes the abstract, the author list, a link to the paper's page, and a direct PDF link when available. It is the fastest way to see what was posted in your field this week, right inside the conversation.

Look up citing papers through Semantic Scholar. Each request returns up to 100 records; the default is 20. This helps trace related research, but the result is not a complete citation census.

Inline references point to papers returned by supported literature tools. Source links and metadata come from those records; external link availability and a paper’s claims are not independently verified by this feature.

Ask Vera to compare papers using the text available in your conversation. The result is model-authored extraction: links help you inspect the papers but do not independently validate each value. Review the table against the original sources, and mark information not supplied by the text as not reported.

When a calculation or simulation spits out a run of integers, Vera checks it against the On-Line Encyclopedia of Integer Sequences and its 370,000+ catalogued sequences. A match instantly tells you whether your mystery numbers have a known closed-form, connect to a published theorem, or show up in a surprising corner of math. Each hit comes with the sequence's name, formulas, and references so you can follow the thread.

Results preserve the provider’s decimal values as text and include source provenance. These are database records, not an independent proof or a machine-verification receipt. Provider access can be unavailable.

Keep any result worth remembering, a summary, a snippet of code, a dataset, a derived equation, in a lasting library tied to your project. Pin any of these anchors into future conversations so Vera carries your best findings forward instead of starting cold each time. Everything is tagged with its verification status, and items Vera created are clearly marked so you always know what is yours versus what was generated.

As you save findings, they automatically link up to related ones by meaning, growing into a web of connected knowledge rather than a flat pile of notes. When you ask a question, Vera can pull not just the closest match but the whole cluster around it, reasoning over related ideas the way a well-read colleague would. One switch sets how far recall reaches: keep it tight to the project at hand, or open it up to connect insights across all of your past work, and Vera widens the net on its own when a question clearly calls for it.

The permission and proof checks for a shared research collection are implemented, but automatic contribution, cross-researcher retrieval, and downstream attribution are not active. Sharing preferences record optional eligibility for future use. Your private research remains excluded, and existing private Project recall is a separate capability.

Browse a curated, ever-changing gallery of open Grand Challenge problems spanning math, physics, biology, and beyond. Each card lays out the problem, the current best-known results, and a single click that drops you into a workspace already primed with the context Vera needs to start real research. As the community cracks existing challenges, fresh frontiers take their place.

Give an agent a research goal and it works the problem for you — searching arXiv, running Python simulations, checking answers with the Z3 solver and Wolfram, and looking up sequences in OEIS and LMFDB. It reasons across multiple steps, then writes up a synthesized report and drops the result straight into your project chat. You can close the tab and come back later; the run keeps going without you.

Watch every running and queued agent in one place, with live timers, a rolling feed of the tools each one is using, and running totals for how many agents finished, how many hit errors, and what it all cost today. The moment an agent completes, its result appears in your chat and the deck updates instantly. It's the fastest way to see what your research fleet is up to at a glance.

See a two-week timeline of your costs and activity, the status of every run with downloadable notebooks, and a breakdown of which tools (Python, Wolfram, Z3, arXiv, Quantum, Lean, and more) you use most and what each one costs. A live feed shows tool activity as it happens, and you can cancel any queued or running agent with one click. Everything updates in real time, so the picture is always current.

The Co-Scientist runs a hypothesis tournament against a Campaign brief and retains the tools, evidence, and outcome. Supported results remain SILVER findings; a generic Lean formalization does not prove the intended natural-language claim or earn GOLD. A finding is progress toward a problem, never a declaration that the Campaign is solved. Runs require owner authorization and a budget; training remains separately gated.

Research, tournament, verification, and peer-review jobs retain their queue state independently of the browser. Workers report progress, cancellation, failure, and completed results to your Project. A restart or ambiguous provider outcome can leave work failed or unresolved; persistence is not a promise of eventual success or automatic safe repetition. Inspect the retained status and cost before starting another attempt.

Durable agent admission limits queued or running work across your projects. Monthly plans change your credit allowance and rate; they do not raise this shared ceiling or establish higher scheduling priority. Enabled capabilities, available funding, and service configuration still determine whether a run can start. If you reach the ceiling, wait for an existing job to finish or cancel work you no longer need.

Instead of nickel-and-diming every step, an agent tallies everything it uses along the way — model tokens, premium tools like Python, Wolfram, Z3, and Lean, and every tool call — and charges you exactly once when it finishes. The charge lands whether the run succeeds or fails partway, so nothing slips through, and the final cost is recorded for you to review. If you bring your own model keys, only the platform-funded tools are billed.

Bring-your-own-key support (your own xAI, OpenAI, Anthropic, or Google account funding the model side) is built and stored securely, currently behind an operator switch while usage-first pricing settles. When enabled, your agents run on your keys with platform-funded tools like Python, Wolfram, Z3, and Lean still metered through your credits.

Talk to Vera and hear her spoken answers as she works, with natural interruptions. The live transcript is a temporary preview in the current tab. Copy any text you want to keep before dismissing the panel or closing the tab.

During a voice call, ask Vera to add a section to your manuscript, search the literature, or launch a research agent. Supported commands use the app's normal permissions and research controls, and results flow back into your chat. Publishing a manuscript remains a separate review and action in the app.

Ask for papers on a topic and Vera searches arXiv, summarizes what she finds in her own voice, and drops the results into your chat. Request a Python simulation mid-conversation and the output appears both spoken and in text. Kick off a background research agent by voice and Vera lets you know the moment it finishes.

While you're on a voice call, Vera stays aware of what's visible — your chat, manuscript, agent results, and main content — so she can answer questions about what's in front of you. She never interrupts to announce it; she only uses it when you ask. You can turn this off in voice settings anytime if you prefer to keep your screen private.

Say something like "add a section on quantum error mitigation to my manuscript" and Vera drafts the text and drops it into your document instantly. The new writing also appears in your chat so nothing is lost. Your manuscript stays in sync with the conversation as it unfolds.

Vera keeps an eye on the research agents running in the background while you talk. When one wraps up, she brings it up naturally in conversation — letting you know the results are ready and offering to summarize them or show you the raw findings. You stay focused on the discussion while your agents work in parallel.

Open the transcript panel to review the latest conversation text, with your turns and Vera's visually distinct. The preview stays in the current tab; copy anything you want to keep before dismissing the panel or closing the tab.

Whenever Vera is speaking and you start talking, she stops on a dime and shifts to listening, just like a person would. There's no awkward lag — interruption is built into the live conversation. It makes talking to Vera feel natural instead of waiting for her to finish.

You can freely mix talking and typing during a voice call. Messages you type — and results Vera pulls up from her tools — are folded into what she knows, so she picks up right where the conversation left off. Voice and text feel like one seamless thread rather than two separate channels.

Platform-funded open-mic calls use server-observed provider connection time, plus each provider-acknowledged text input. Research tools have their own metered usage. Stopping closes the paid connection; uncertain provider usage requires reconciliation before another funded call. Browser estimates are not billing authority, and provider-funded calls use your own provider account.

Tap the microphone in the chat bar and your speech turns into text in real time as you talk, so you can see it forming. Stop, and the finished text lands in the box ready for you to review and send. It's a fast way to compose messages hands-free without starting a full voice call.

Pick the voice that feels right to you and it sticks across your conversations. Eve is the default, and you can switch anytime. Every voice carries the same intelligence and responsiveness — only the tone changes.

Synthesize the Project's research conversation into a structured draft with mathematics and a LaTeX source. Review the generated text and citations before submission. Any model-written pre-audit is commentary, not canonical Sigma evidence or a verification receipt; current archive Sigma comes from the persisted evidence pipeline.

Publish a finished manuscript to the Vanyx Science Archive with an abstract, subject tags, date, and available Sigma evidence. The article is attributed to your account while that link exists. Account deletion can remove an article or preserve it without your account link when needed to retain another person’s contribution; see the Privacy Policy for retention limits.

Sigma is computed from evidence rather than a model's novelty prediction. The public archive reads the stored score, standing, evidence kinds, and engine identity after checking publication provenance and the current Project evidence generation. Missing, inconsistent, or legacy-only evidence displays Verification unavailable. Standing can be refuted or contested even with a numeric score; Sigma is not a probability that a paper is true.

Patent Studio drafts the title, background, summary, detailed description, and proposed claims. Export available PDF or LaTeX artifacts for review. Generated wording, formatting, or a model pre-audit does not establish legal sufficiency, patentability, a filing, or a priority date; those require your filing process and qualified review.

The preview formats the draft and mathematical expressions as printable HTML. Use your browser’s print controls to print or save a PDF. Review the content and formatting before sharing; use the separate LaTeX source export for external editing or typesetting.

Export the generated source as a .tex file. Review and compile it in your chosen typesetting environment, and check the requirements of the destination before submission. Patent exports remain drafts for qualified review; exporting a file does not establish compilation, filing, or acceptance.

Turn your manuscript into a tidy, print-styled page that opens straight to your browser's print or save-as-PDF dialog. It is the fast, lightweight way to share a draft or archive your work when you don't need full typesetting.

The public archive shows titles, abstracts, topic tags, author, date, and canonical Sigma evidence when available. Missing or stale evidence is labeled unavailable and cannot pass a positive Sigma threshold. Publication alone does not certify a claim. Open a manuscript to read its prose and rendered equations.

Signed-in researchers can leave a star rating and written assessment. Available reviews show the author’s public display name, timestamp, and rating. Eligibility checks and rate limits restrict submissions; a review is a human opinion, not formal verification or a guarantee of spam-free feedback.

Describe a scene and Vera renders a short video for you — opened from the same create menu as manuscripts and surfaces. The clip is saved to your project and downloadable. Metered from your balance like every other tool.

Campaigns connect a problem brief, community, contributions, and retained findings. Each result keeps its actual standing: tool-supported hypotheses and draft manuscripts do not become machine-verified proofs merely by appearing in a Campaign. Joining opens a workspace with the Campaign context; formal proof authority requires the current exact-statement verification workflow.

Campaign scores count selected public formal proofs at 20 points each. Global scores also count public Archive publications at 10 points each. Private Projects, drafts, memberships, and anchors are excluded. Public proof credit requires exact current-policy GOLD evidence; publication credit does not imply verification. The score is not a probability of correctness or scientific quality.

The researcher page shows public formal Campaign proofs and public Archive publication totals, using the same public-only projection as the boards. Private activity is excluded. Proof credit requires current-policy GOLD evidence; publication credit does not certify the manuscript, guarantee reputation, or establish downstream influence.

The private Impact page displays existing contribution receipts. Automatic sharing, downstream reuse, and new attribution receipts are not active. Campaign contribution scores already count eligible work separately; a score does not establish downstream influence or create a payment entitlement.

Once a paper is published to the global archive, any signed-in researcher can leave a star rating and written assessment. Reviews go live immediately with no approval gate—each appears publicly with the reviewer's name, timestamp, and rating. You can review each paper once, and light rate limits keep the feed free of spam.

Every Grand Challenge comes with a dedicated community where members chat, coordinate, and share findings, and the Vanyx Commons Town Square sits above them all for cross-frontier discussion. A searchable directory lists every public community with its name, description, and member count, so you can find one around your interests and join in a single click.

Sign in, name your community, add a description, and choose public (anyone can join) or private (invite only). It's ready instantly with a general text channel and a voice lounge already set up, and you're the owner with full moderator control. Public communities appear in the searchable directory; private ones stay hidden and open only to people you invite—ideal for labs, sensitive research, or closed teams.

Owners and moderators can add as many text and voice channels as they need, and mark any of them private so only selected members can read and post. This lets a single community hold open discussion, topic-specific research spaces, and closed admin coordination without ever splitting into separate servers.

Community channels update instantly—messages appear the moment they're sent, and typing indicators show who is composing right now. Mention @vera to summon Vera Co-Scientist directly in the channel, where her answer posts as a reply visible to the whole group. History loads smoothly as you scroll back, and authors and moderators can remove their own or others' messages.

Community owners and moderators can remove any message instantly to keep channels focused and on-topic, without having to ban whole accounts. On top of that, automatic moderation screens every message for spam, slurs, and scam patterns and blocks offending posts before they ever appear, with a brief reason shown to the sender.

Each voice channel shows a live stage with everyone currently present, displayed by name alongside a chat sidebar so listeners can comment as speakers talk. Joining instantly puts your name in front of the group, and you can step back to text-only chat anytime. Live audio is the next phase.

Each community shows a roster of its members with names, join dates, and roles such as owner, moderator, or member. The list refreshes in real time as new people arrive, and clicking any member jumps straight to their researcher profile.

Your joined communities line up in a side rail ordered by recent activity, with the Vanyx Commons Town Square always pinned first. Click any one to jump straight into its channels and members. Leave a community to remove it from the rail, and rejoin whenever you like.

Joining records your membership without publishing your private chats or research. Eligible public formal proofs can count toward the Campaign leaderboard; membership alone does not earn contribution credit.

Status badges describe the Campaign’s workflow. They do not establish that its scientific question is unsolved, partially solved, or solved; assess the underlying evidence and proof receipts.

A bounded welcome grant helps you try Vera without a card. Add credits with a one-time purchase when you need more; Billing shows your balance and recorded usage. Hosted service fees are separate from additional credit purchases. Availability still depends on rollout, model access and configured services; automatic top-ups are not available yet.

The new Researcher and Lab offers combine a hosted research service fee with an included compute grant. Their terms are listed on Pricing, and they are not available for purchase yet. Existing paid terms and banked credits are preserved. Available research capabilities and verification standards do not depend on a service plan.

Add a public HTTPS endpoint and optional key in settings. Vanyx resolves and tests the connection before saving it securely, then exposes the model in your picker. Localhost and private-network addresses stay blocked because the hosted Vanyx server cannot safely reach a user's laptop. Self-hosted Vanyx retains the Ollama path; arbitrary-user local models need the planned device bridge before we can honestly offer them on the public web.

Create, rename, archive, and delete projects, and sort them by date or filter down to the ones that led to a published paper or patent. Full-text search finds anything by name across all your work. Your dashboard shows recent projects at a glance with document counts and dates.

Set a public display name, pick a custom accent color that instantly recolors the whole interface, and upload your own logo (up to 2MB) to replace the Vanyx wordmark — turning the workspace into a branded lab. Every change applies live, the moment you make it. Like everything here, it's included for everyone.

Billing shows your prepaid balance, current credit rate, plan, and recorded payments. Make a one-time credit purchase, update your saved card, or choose Stop future charges to request cancellation. Card and cancellation requests show confirmed, pending, or unavailable states. The separate Stripe link opens invoice history. Automatic top-ups are not available yet; complimentary and admin accounts are clearly identified.

Account settings control training and sharing consent, password changes, and supported data-erasure requests. Account deletion coordinates owned application data and active billing. Legal, security, and financial retention obligations and external-provider disposition still apply; read the Privacy Policy for the current scope and limits. Training consent does not enable a training stage.

Land on Vanyx and start chatting right away as a guest — no email needed. When you ask your first real question or start a project, a single email box appears that either signs you in as a returning user or creates a new account on the spot, with your work carried over seamlessly. Guests can browse public challenges, communities, and papers, and sign in whenever they want to post or join.

Press ⌘K (or Ctrl+K) and type. Anything you write can start a new chat instantly ("solve: …"), or pick from the list to open a workbench app or jump to any page. Pick an app before you even open a chat and it's waiting for you when the workspace loads. Built keyboard-first, like a professional instrument should be.

The left rail hosts your workbench: click an icon and that app docks next to the conversation; drag the divider to resize, or pop any app out into its own draggable, resizable floating window. On your phone the same apps live in a real bottom tab bar built for thumbs, not a shrunken desktop.

Where your browser supports installation, open Vanyx as a standalone app. A network connection is needed for live research, account actions and synchronization. Installation and background behavior depend on your browser and device.

Anywhere you see a question-mark icon, click it for a plain-language explanation of what that control does — what it changes, what it costs, and when you'd use it. Explanations appear in place without leaving the page, and you can toggle the icons on or off globally whenever you want a cleaner interface.

Forgot your password? Enter your email on the login page and we'll send a secure link to set a new one. Already signed in? Update your password anytime from your account settings. Both flows are protected against abuse and delivered by email.

New to a research platform this deep? Toggle Explain-everything mode from the rail or Settings and small info badges bloom on controls throughout the app — the workbench apps, the command palette, your live dashboard stats, and more. Hover or tap any badge for a clear, jargon-free description of exactly what that control does and why you'd use it. Turn it off and every badge vanishes — zero clutter when you don't need it. It remembers your choice.

As a conversation grows, the blackboards and interactive surfaces Vera opens scroll up with the transcript. The Active Tools tray keeps a running index pinned in view: click a chip to scroll a board back into focus or pop it out into a floating window, and click a surface to bring its window forward. The same live index appears in Mission Control, so nothing you're working with is ever more than one click away.

Instead of a blank “ask me anything,” the chat box animates through concrete examples of what the system can actually do — proving there are infinitely many primes, simulating a damped pendulum, verifying a theorem in Lean, searching 220M papers, running a hypothesis tournament — typed out and cycled one after another. Start typing and it steps aside instantly.

Research Workspace collapses the work into seven focused tabs: Mission (the dossier — scorecard, distance-to-summit, mission graph, and a Library that opens papers, notebooks, datasets, and connections as floating windows), Evidence (private human-authored Research Objects with immutable revisions, typed relations, evidence-gated standing changes, and preserved negative results), Lab notebook, Writing, Proofs, Anchors, and Agents/Runs. Any durable user or Vera message can enter an explicit review that preserves its source provenance and still starts at OPEN / rung 0. Vera can open any tab through open_panel, but opening Evidence grants no write or verification authority.