{"generatedAt":"2026-09-22T07:56:58.671Z","source":"https://opencharts.com/models","models":[{"slug":"moonlight","url":"https://opencharts.com/models/moonlight","publicName":"Theo Moonlight 1.1","shortName":"Theo Moonlight","version":"1.1","kind":"image","kindLabel":"Image engine","status":"live","statusLabel":"Live in Theo","releasedAt":"2026-09-15","tagline":"High fidelity image renders on our own hardware.","summary":"Theo Moonlight is the image engine OpenCharts runs itself: Theo's own text-to-image engine, served from hardware we operate, with a prompt screen before every render and an output classifier after it. It lives inside Theo Stealth, where fidelity is the whole point, and it is the one engine there that carries the Theo name because it is the one we run.","hero":{"url":"https://opencharts.com/models/moonlight-hero.jpg","alt":"A thin crescent of glowing silver light suspended above still black water, reflected as a clean vertical streak, with violet and indigo light bleeding into the dark around it.","caption":"Rendered on Theo Moonlight 1.1, on our own hardware, at 1344 by 768."},"whatItDoes":["Renders about one megapixel from a written prompt, at seven aspect ratios from square to 16:9.","Honours a negative prompt, so a do-not-do-that instruction genuinely steers the picture.","Stays the same between renders. The engine changes only in a deliberate, versioned release, never by drift.","Is picked by name in the Stealth Engine menu and never silently swapped for a partner engine. If our hardware is down, you see an error, not a quiet reroute."],"howItWorks":[{"title":"Your prompt is screened first","body":"Before anything is rendered, the prompt passes a screen that reads the whole request rather than matching a word list. A refusal costs nothing and reads the same whatever the reason."},{"title":"The render runs on our hardware","body":"Dedicated hardware that OpenCharts operates produces the picture at full quality, at a resolution chosen for the aspect ratio you asked for. Nothing about the render leaves hardware we run."},{"title":"Every image is classified before you see it","body":"An output classifier checks the result. It fails closed: if the detector cannot answer, the render is refused rather than shown."},{"title":"The render is kept briefly for review","body":"Because we run Moonlight ourselves, the prompt and the image are retained for a limited time so a small team can confirm the controls are working. Never used for training, never sold or shared for advertising, access restricted and logged."}],"measured":[{"label":"Render time","value":"33 seconds","detail":"A 1024 by 1024 image at the default quality setting","measuredOn":"2026-09-15","source":"Measured on our own hardware"},{"label":"Output","value":"About 1 megapixel","detail":"Seven aspect ratios from 1:1 to 16:9, each around one million pixels","measuredOn":"2026-09-15","source":"Engine configuration"},{"label":"Safety review round","value":"11 of 12 verdicts correct","detail":"Zero misses and one false positive on the output classifier in the latest human review","measuredOn":"2026-09-16","source":"Human review of a labelled render corpus"}],"safety":["A prompt screen runs before every render, including prompts an assistant rewrote, so a rewriter cannot launder a refused prompt.","An output classifier runs after every render and fails closed: no answer from the detector means no image.","The screen blocks the intersection of sexual and apparently-minor cues. A lawful adult prompt is not blocked for brushing a word.","One refusal message for every reason, so the control cannot be probed by watching which message comes back.","Refused requests are recorded against the account, without the prompt.","Renders and prompts are kept for a limited time for review. Material connected to a child-safety report is kept longer because the law requires it."],"whereItRuns":[{"label":"Hardware","value":"Dedicated hardware that OpenCharts operates"},{"label":"Versioning","value":"The engine changes only in a deliberate release, never by drift"},{"label":"Your prompt","value":"Never sent to a third-party model provider"},{"label":"Retention","value":"Prompt and image kept briefly for safety review, then deleted"}],"availability":{"surface":"Theo Stealth","url":"https://opencharts.com/chat","note":"Open Theo Stealth and pick Theo Moonlight in the Engine menu. It is pinned to the top whenever our hardware is answering, and omitted, never greyed, when it is not.","helpUrl":"https://opencharts.com/help/theo-stealth"},"benchmarksUrl":null,"distribution":[{"channel":"Inside Theo","status":"live","statusLabel":"Live","note":"In Theo Stealth today.","url":"https://opencharts.com/chat"},{"channel":"Public API","status":"planned","statusLabel":"Planned","note":"A metered endpoint with the same prompt screen and output classifier in front of it.","url":null},{"channel":"Third-party arenas","status":"planned","statusLabel":"Planned","note":"Follows the public API: arenas rank publicly callable models.","url":null}],"faq":[{"question":"Why is Theo Moonlight only in Theo Stealth?","answer":"Stealth is the surface built for fidelity, where the slowest engine we offer is a fair trade for the most control. Moonlight is that engine: our own hardware, full-quality renders, a versioned engine that changes only on release. Everywhere else, Theo orchestrates faster lab engines and shows which one answered."},{"question":"What makes Moonlight one of our own models?","answer":"It runs end to end on hardware OpenCharts operates. The render pipeline, the prompt screen, the output classifier, the negative-prompt handling and the retention policy are ours, no part of a render leaves hardware we run, and the engine is versioned by us and changes only when we release it."},{"question":"Why does Moonlight keep my prompt when the rest of Stealth keeps nothing?","answer":"Because we run it ourselves, we are directly accountable for what it makes. Every Moonlight render is retained with its prompt for a limited time so a small team can confirm the safety controls are working. It is never used for training, never sold, and the picker names Moonlight clearly so the choice is yours."},{"question":"Why is it the slowest image engine you offer?","answer":"Moonlight renders at full quality on our own hardware, about 33 seconds for a megapixel. That is the cost of fidelity and of running the engine ourselves, and it is deliberate: Moonlight is the engine where fidelity is the entire point."}]},{"slug":"real-world","url":"https://opencharts.com/models/real-world","publicName":"Theo Real World 1.1","shortName":"Theo Real World","version":"1.1","kind":"capture","kindLabel":"Capture engine","status":"rolling-out","statusLabel":"Rolling out","releasedAt":"2026-09-21","tagline":"Photos of a real thing become a 3D capture of the thing in its space.","summary":"Theo Real World is Theo's own capture engine. Photos, or a walk-around video, of one real person, object or space become a faithful 3D capture: the cameras are solved from the pictures themselves and the capture is trained against the actual pixels. Nothing is invented. Coverage decides quality, and the card says which sides were seen.","hero":null,"whatItDoes":["Takes up to 60 photos or one walk-around video of one subject and returns a capture you can orbit, walk through and view in AR.","Solves where every shot was taken from the pixels themselves. No angle labels, no guessing.","Delivers two tiers on one card: a quick capture in about a minute, a full capture a few minutes later on the same row.","Says which sides were seen. A capture is a reconstruction, never a guess: gaps stay empty and the card names them."],"howItWorks":[{"title":"Frames are chosen","body":"24 to 48 sharp, evenly spread frames are picked from your photos or your clip, so the capture is built from the clearest views you gave it."},{"title":"The cameras are solved","body":"Where every shot was taken is recovered from the pictures themselves, in about a second, and the scene is aligned so the subject sits at the centre."},{"title":"The capture is trained against your pixels","body":"The 3D capture is optimised to reproduce your photographs, view by view. The quick tier is published mid-run; the full tier follows on the same job."},{"title":"The capture ships small","body":"Both tiers are written in a compact format, so a full capture is about 10 MB and loads in seconds, with the solved cameras beside it so the viewer starts where your first photo was taken."}],"measured":[{"label":"Quick capture","value":"127 seconds","detail":"From submit, on a 9 second walk-around clip; 40 camera positions solved from 120 candidate frames","measuredOn":"2026-09-21","source":"Measured end to end on the production video path"},{"label":"Full capture","value":"269 seconds","detail":"The same clip, end to end; a 10.5 MB capture that reproduces held-out frames faithfully","measuredOn":"2026-09-21","source":"Measured end to end on the production video path"},{"label":"From nine stills","value":"59 s quick, 157 s full","detail":"On a warm worker; 4.4 MB and 10.7 MB files","measuredOn":"2026-09-20","source":"Measured on our capture hardware"},{"label":"Camera solve","value":"1.1 seconds","detail":"Every camera's position for the nine-still set","measuredOn":"2026-09-20","source":"Measured on our capture hardware"}],"safety":["A capture shows what the cameras saw. Nothing is filled in, and the coverage line names the sides that were not seen.","Your photos and clips stay inside the platform's own storage; the capture is written back to your own account.","Credits are charged once at dispatch. A failed full tier keeps the quick capture and is never re-billed."],"whereItRuns":[{"label":"Hardware","value":"Dedicated hardware that OpenCharts operates"},{"label":"Your media","value":"Stays inside the platform's own storage; nothing is sent to a third-party model provider"},{"label":"Versioning","value":"The engine changes only in a deliberate release, never by drift"}],"availability":{"surface":"Theo Chat, Real World mode","url":"https://opencharts.com/chat","note":"Rolling out. Where it is enabled, pick Real World in the Create row of the mode selector, add photos or a walk-around video, and press Capture now.","helpUrl":"https://opencharts.com/help/real-world"},"benchmarksUrl":null,"distribution":[{"channel":"Inside Theo","status":"rolling-out","statusLabel":"Rolling out","note":"The Real World mode in Theo Chat, where enabled.","url":null},{"channel":"Third-party evaluation","status":"planned","statusLabel":"Planned","note":"Reconstruction quality is measured against held-out frames today; a public evaluation follows.","url":null}],"faq":[{"question":"How is this different from Theo 3D?","answer":"3D invents; Real World captures. Theo 3D builds an object from a description or a few views. Real World reconstructs something that exists, from the photographs you took, and shows only what the cameras saw."},{"question":"Why does a walk-around video work better than photos?","answer":"Coverage decides quality. A short clip in which the camera moves around the subject gives the engine dozens of viewpoints; eight stills of a small glossy object are the hardest case for any capture. The capture says which sides were seen either way."},{"question":"Is anything in a capture generated?","answer":"No. The capture is trained to reproduce your pixels, and a side no camera saw stays empty. The card names those sides rather than filling them in."}]},{"slug":"theoarca-magnus","url":"https://opencharts.com/models/theoarca-magnus","publicName":"TheoArca Magnus 5.1","shortName":"TheoArca Magnus","version":"5.1","kind":"text","kindLabel":"Text engine","status":"rolling-out","statusLabel":"Rolling out","releasedAt":"2026-07-25","tagline":"Theo's own deep-reasoning engine, governed and audited.","summary":"Theo's own engines, built and operated by TheoVex. TheoArca Velox answers fast; TheoArca Magnus reasons deeply. Both reach Theo through Theo's orchestration API on a key routed only to Arca. Magnus is the deep-reasoning tier: it reasons at full effort by default and answers in text.","hero":null,"whatItDoes":["Reasons deeply by default, with a 1M-token context window and answers of up to 131K tokens.","Passes every request through a privacy guardrail: sensitive values are tokenized before inference and prompt-injection attempts are blocked.","Records every request on an append-only, hash-chained audit ledger, tagged with its purpose.","Answers in text. Flowcharts, notes, images and files are built by Theo's usual engine."],"howItWorks":[{"title":"You pick it, or Theo does","body":"Choose Theo Arca from the composer, pin it as your engine in Settings, or pick a compliance tier in Code Canvas. Theo keeps its usual engines on standby behind it."},{"title":"The request is screened and tagged","body":"Sensitive values are tokenized before inference, prompt-injection attempts are blocked, and the request is tagged with the purpose it serves."},{"title":"The engine answers in text","body":"Flowcharts, notes, images and files are built by Theo's usual engine; Arca answers in words, with a 1M-token context window."},{"title":"The ledger records it","body":"Every request lands on an append-only, hash-chained audit ledger, so what ran and why can be shown after the fact."}],"measured":[{"label":"Context window","value":"1,000,000 tokens","detail":"The whole conversation, documents included, in one request","measuredOn":"2026-09-03","source":"Gateway limits, read from the live endpoint"},{"label":"Longest answer","value":"131,072 tokens","detail":"The gateway's output ceiling per request","measuredOn":"2026-09-03","source":"Gateway limits, read from the live endpoint"}],"safety":["A privacy guardrail screens every request: sensitive values are tokenized before inference and prompt-injection attempts are blocked.","Every request lands on an append-only, hash-chained audit ledger, tagged with its purpose.","Zero retention: nothing you send is used to train anything.","Compliance is the product name for the governed pipeline. It is not a HIPAA or BAA claim; the managed deployment is the non-PHI tier."],"whereItRuns":[{"label":"Who runs it","value":"TheoVex, Theo's own lab, serves the TheoArca 5.1 family behind a governed gateway."},{"label":"How it's reached","value":"Through Theo's orchestration API (hitheo.ai), on a dedicated key whose routing is pinned to Arca."},{"label":"Your data","value":"Zero retention. Nothing you send is used to train anything."},{"label":"Guardrail","value":"A privacy guardrail screens every request: sensitive values are tokenized before inference and prompt-injection attempts are blocked."},{"label":"Audit","value":"Every request lands on an append-only, hash-chained audit ledger, tagged with its purpose."},{"label":"Capacity","value":"A 1M-token context window and answers of up to 131K tokens."},{"label":"Answers in text","value":"Flowcharts, notes, images and files are built by Theo's usual engine, Arca answers in words."}],"availability":{"surface":"Theo Chat and Code Canvas","url":"https://opencharts.com/chat","note":"Available where enabled: choose Theo Arca in the composer, pin it in Settings, or pick the deep-reasoning compliance tier in Code Canvas.","helpUrl":"https://opencharts.com/help/engines"},"benchmarksUrl":"https://opencharts.com/benchmarks/theoarca-magnus-5-1","distribution":[{"channel":"Inside Theo","status":"rolling-out","statusLabel":"Rolling out","note":"The composer's Theo Arca pick, an engine pin in Settings and the Code Canvas compliance tiers, where enabled.","url":null},{"channel":"Public API","status":"in-progress","statusLabel":"In progress","note":"Served through the TheoVex gateway today; a public, metered listing on model marketplaces is in progress.","url":null},{"channel":"Third-party evaluation","status":"planned","statusLabel":"Planned","note":"Follows the public listing. Until then the benchmarks page shows OpenCharts' own runs, labelled internal.","url":null}],"faq":[{"question":"Is this the engine Theo usually answers with?","answer":"No. Let Theo orchestrate runs the standard lab engines and shows which one answered each step. TheoArca is the composer's pick and a pin: it leads only when you choose it, and Theo's usual engines stay on standby behind it."},{"question":"What does governed mean?","answer":"Every request passes a privacy guardrail that tokenizes sensitive values and blocks prompt-injection attempts, is tagged with its purpose, and is recorded on an append-only, hash-chained audit ledger. These are mechanisms the platform documents, not certifications: it is not a HIPAA or BAA claim."},{"question":"Where are its benchmark scores?","answer":"On the benchmarks page, labelled internal until a third party publishes a score. OpenCharts runs the public tests itself with versioned prompts and deterministic graders, records the engine that actually served each sample, and refuses to write a score the requested engine did not serve."}]},{"slug":"theoarca-velox","url":"https://opencharts.com/models/theoarca-velox","publicName":"TheoArca Velox 5.1","shortName":"TheoArca Velox","version":"5.1","kind":"text","kindLabel":"Text engine","status":"rolling-out","statusLabel":"Rolling out","releasedAt":"2026-07-25","tagline":"Theo's own low-latency engine, governed and audited.","summary":"Theo's own engines, built and operated by TheoVex. TheoArca Velox answers fast; TheoArca Magnus reasons deeply. Both reach Theo through Theo's orchestration API on a key routed only to Arca. Velox is the fast tier: quick, focused answers on the same governed pipeline.","hero":null,"whatItDoes":["Answers fast, with the same 1M-token context window and the same 131K-token answer ceiling as Magnus.","Passes every request through a privacy guardrail: sensitive values are tokenized before inference and prompt-injection attempts are blocked.","Records every request on an append-only, hash-chained audit ledger, tagged with its purpose.","Answers in text. Flowcharts, notes, images and files are built by Theo's usual engine."],"howItWorks":[{"title":"You pick it, or Theo does","body":"Choose Theo Arca from the composer, pin it as your engine in Settings, or pick a compliance tier in Code Canvas. Theo keeps its usual engines on standby behind it."},{"title":"The request is screened and tagged","body":"Sensitive values are tokenized before inference, prompt-injection attempts are blocked, and the request is tagged with the purpose it serves."},{"title":"The engine answers in text","body":"Flowcharts, notes, images and files are built by Theo's usual engine; Arca answers in words, with a 1M-token context window."},{"title":"The ledger records it","body":"Every request lands on an append-only, hash-chained audit ledger, so what ran and why can be shown after the fact."}],"measured":[{"label":"Context window","value":"1,000,000 tokens","detail":"The whole conversation, documents included, in one request","measuredOn":"2026-09-03","source":"Gateway limits, read from the live endpoint"},{"label":"Longest answer","value":"131,072 tokens","detail":"The gateway's output ceiling per request","measuredOn":"2026-09-03","source":"Gateway limits, read from the live endpoint"}],"safety":["A privacy guardrail screens every request: sensitive values are tokenized before inference and prompt-injection attempts are blocked.","Every request lands on an append-only, hash-chained audit ledger, tagged with its purpose.","Zero retention: nothing you send is used to train anything.","Compliance is the product name for the governed pipeline. It is not a HIPAA or BAA claim; the managed deployment is the non-PHI tier."],"whereItRuns":[{"label":"Who runs it","value":"TheoVex, Theo's own lab, serves the TheoArca 5.1 family behind a governed gateway."},{"label":"How it's reached","value":"Through Theo's orchestration API (hitheo.ai), on a dedicated key whose routing is pinned to Arca."},{"label":"Your data","value":"Zero retention. Nothing you send is used to train anything."},{"label":"Guardrail","value":"A privacy guardrail screens every request: sensitive values are tokenized before inference and prompt-injection attempts are blocked."},{"label":"Audit","value":"Every request lands on an append-only, hash-chained audit ledger, tagged with its purpose."},{"label":"Capacity","value":"A 1M-token context window and answers of up to 131K tokens."},{"label":"Answers in text","value":"Flowcharts, notes, images and files are built by Theo's usual engine, Arca answers in words."}],"availability":{"surface":"Theo Chat and Code Canvas","url":"https://opencharts.com/chat","note":"Available where enabled: choose Theo Arca in the composer, pin it in Settings, or pick the low-latency compliance tier in Code Canvas.","helpUrl":"https://opencharts.com/help/engines"},"benchmarksUrl":"https://opencharts.com/benchmarks/theoarca-velox-5-1","distribution":[{"channel":"Inside Theo","status":"rolling-out","statusLabel":"Rolling out","note":"The composer's Theo Arca pick, an engine pin in Settings and the Code Canvas compliance tiers, where enabled.","url":null},{"channel":"Public API","status":"in-progress","statusLabel":"In progress","note":"Served through the TheoVex gateway today; a public, metered listing on model marketplaces is in progress.","url":null},{"channel":"Third-party evaluation","status":"planned","statusLabel":"Planned","note":"Follows the public listing. Until then the benchmarks page shows OpenCharts' own runs, labelled internal.","url":null}],"faq":[{"question":"Is this the engine Theo usually answers with?","answer":"No. Let Theo orchestrate runs the standard lab engines and shows which one answered each step. TheoArca is the composer's pick and a pin: it leads only when you choose it, and Theo's usual engines stay on standby behind it."},{"question":"What does governed mean?","answer":"Every request passes a privacy guardrail that tokenizes sensitive values and blocks prompt-injection attempts, is tagged with its purpose, and is recorded on an append-only, hash-chained audit ledger. These are mechanisms the platform documents, not certifications: it is not a HIPAA or BAA claim."},{"question":"Where are its benchmark scores?","answer":"On the benchmarks page, labelled internal until a third party publishes a score. OpenCharts runs the public tests itself with versioned prompts and deterministic graders, records the engine that actually served each sample, and refuses to write a score the requested engine did not serve."}]}]}