One console, three systems, WEB PORTAL (API ONLY, online, you are here) · PC LOCAL (full admin console on your PC, GPU when present) · USB LOCAL (stand-alone stick agent, plug and play, mobile). Same limbs, same P0 gate, same labels; only the brain differs.
A ninja console with no GPU: bring a key from any vendor (Groq, Gemini, OpenRouter, OpenAI, Grok, DeepSeek, …). Pick the provider → paste the key → Connect → chat. Keys stay in this tab, chatagent.ca never sees them. Browser tools + 15 champions are already installed on the page. Want it all offline on your own machine? Local GGUF console, same V1 bench, no API key at all. Legal: legal & disclaimers. Steward: Justin Helmer.
API portal · pick a provider → paste its key → Connect
models, the list is read from your key, not a hard-coded table
Paste it in the key box. Click Connect. The model is chosen for you.
Then chat. Skills and champions are already on this page, toggle them in the left rail.
Optional: press ⇄ Compare to put one prompt to two or more models at once and read the answers side by side. Then press the + on the strongest segments of each answer and Assemble them into one, or leave nothing picked and press ⇄ Synthesize for one refined answer.
⇄ Compare benchtwo lanes, one prompt, side by side, then one refined answer
Put the same prompt to two or more models at once, then read them side by side. From there, one click turns the comparison into something you can use: press the + on the strongest segments of each answer and Assemble them, your own picks, in order, verbatim, tagged to the answer each came from and written without spending a call, or leave nothing picked and press ⇄ Synthesize to send every answer to one chosen lane and get a single refined answer back, with the line saying what it leaned on. If you have picked segments, Synthesize sends only those, so the fuse is sharp and costs less. One answer per run joins the conversation, so the log stays a conversation. Keys pasted in a lane stay in this tab’s memory, never in browser storage; nothing here is uploaded and no answer is stored anywhere but your own browser. A lane set to the same provider as the top box shares the key you already pasted there. A custom lane can point at any OpenAI-compatible URL (two self-hosted servers, two labs), and 127.0.0.1/localhost is allowed by this page’s own policy. Each run spends one call per ready lane on your keys (the synthesis spends one more), and the seconds and tokens of each answer are shown with it.
Image generation
Type a prompt, choose where the picture is made, and it appears right here in the module. Nothing is
uploaded to chatagent.ca: a picture is either made on a LYGO console you attach (the same SDXL engine the PC LOCAL
console uses) or by the key you pasted above, that key stays in this tab.
Loading routes…
Which routes this page can really use, and what each one costs
A console draws one picture at a time (3 per 10 minutes per address): seconds on a free card, about
5 minutes on the CPU route the engine picks while the chat model holds the card. A key route bills your own
account, usually free or a few cents a picture.
This page has no GPU and no model. Install one engine on your own machine — image, sound or
video — and this module drives it through a small local bridge. Nothing is uploaded, and the portal cannot reach
it unless it is running. Build a limb → · or take the
full console if you want all three plus the chat engine.
A limb answering as stub proves the wiring, not the model — /health always names
the backend. Ports: image 8191 · sound 8192 · video 8193, all on 127.0.0.1.
Standalone image portal
ChatAgent Image Studio
Full-size BYOK image and video studio. Every provider gets a read-only preflight check before generation; keys stay in this browser. Existing portal tools are untouched.
This page (API portal). Pick a provider, paste that vendor’s key, Connect, then chat. Groq is one free start, not the only one. Keys stay in your browser. chatagent.ca never sees the key.
Limits change. Prefer the provider’s own page. This portal accepts any OpenAI-compatible key: pick the vendor or choose Any OpenAI-compatible URL, paste base + key + model id. Keys stay in this tab.
Unknown vendor? Choose Any OpenAI-compatible URL, paste their base (we append /v1/chat/completions if needed), key, and model id. Directories: freellm.net · awesome-freellm-apis.
Disclaimers
RESOURCE, not CANON. Dual ledgers / Haven Star Chart remain canon, nothing here writes to them.
No account. Your API key stays in this browser and goes only to the vendor you chose; chatagent.ca servers never see it.
Not medical, legal, or financial advice. Models can be confidently wrong, check anything that matters. You remain the publisher of what you send and what you do with the answer.
Your key, your vendor, their terms: pricing, quotas, retention and acceptable use are between you and that vendor. Vendor names are their owners' marks; no affiliation or endorsement is claimed. Key can spend real money.
No warranty. Provided “as is”, no uptime promise. This domain does not host a public GPU. CORS failures are the vendor blocking browsers; try Groq/OpenRouter or the local kit.
Models keep their authors' licences (Gemma and Qwen are Apache-2.0; llama.cpp is MIT). The console code is LYGO Sovereign License v3.0, source-available, not MIT: use it, build on it, do not sell, rebrand or strip attribution.
