{"ok":true,"service":"agents-api","version":"1.0.0","pay_to":"0x1D15aC2CAa30aBF43D45cE86eE0cB0F3C8B929F6","network":"eip155:8453","asset":"USDC","mock_mode":false,"contact":"maximusmediahouse@agentmail.to","delivery":{"durable_content_addressed":true,"durable_url_template":"https://agent.maximusmediahouse.com/v1/artifacts/{sha256}","signed_url_ttl_s":86400},"host":"127.0.0.1","port":8020,"public":true,"public_port":false,"exposed_via":"mtls_tunnel","public_base_url":"https://agent.maximusmediahouse.com","chat_model_resident":"qwen-local","caveats":["flux3-klein RE-LISTED 2026-10-28 after the route was fixed and proven: the live ComfyUI was bumped to v0.41.1 (adds the 'flux3' clip type and the flux3 model detector) and the flux3 graph's start latent switched to EmptyFlux2LatentImage (FLUX.3 in_channels=128). A real 1024px render verified on :8002 — see /health SOLD.","flux-2-klein-4b (fast): TEXT renders unreliably — phrases >=2 words garble ~2/3 of the time; short single words survive. Klein runs ~6s warm (~133s cold), NOT sub-second.","qwen3.5 chat is served by the LOCAL llama-server; billed FLAT PER CALL ($0.003/call, not per token) since 2026-10-28.","hunyuan-480p-i2v video + prepaid top-up are UNAVAILABLE — NOT SOLD (both return 503 before any payment gate). The arena's live GPU render is not covered by the house one-GPU lock, so long GPU jobs can be killed mid-flight by a lane OOM + supervisor restart.","UNAVAILABLE — NOT SOLD: hunyuan-480p-i2v is withdrawn from the catalogue (2026-10-28). The route returns 503 before any payment gate. Reason: there is no real serialisation between the live arena GPU render and paid GPU work (the house one-GPU lock was FREE while arena-renderer.service rendered at ~99% GPU), so a ~5-10 minute clip can be killed mid-flight by a lane OOM + supervisor auto-restart. Not re-listed until the arena participates in the lock."],"videos":{"available":false,"sold":false,"endpoint":"https://agent.maximusmediahouse.com/v1/videos/generations","queue":"https://agent.maximusmediahouse.com/v1/videos/queue","reason":"hunyuan-480p-i2v is UNAVAILABLE (2026-10-28). MEASURED: there is no real serialisation between the live arena GPU render and paid GPU work (the house GPU lock was free while arena-renderer.service rendered at ~99% GPU), so a multi-minute video render can be killed mid-flight by a lane OOM + supervisor restart. Not sold until that is fixed."},"unavailable":{"hunyuan-480p-i2v":"hunyuan-480p-i2v is UNAVAILABLE (2026-10-28). MEASURED: there is no real serialisation between the live arena GPU render and paid GPU work (the house GPU lock was free while arena-renderer.service rendered at ~99% GPU), so a multi-minute video render can be killed mid-flight by a lane OOM + supervisor restart. Not sold until that is fixed."},"models":{"flux-2-klein-4b":{"id":"flux-2-klein-4b","pool":"halo:8002","label":"FLUX.2 klein 4B fp8 (fast, cheap)","description":"FLUX.2 klein 4B fp8 text-to-image — fast, cheap prototype/iteration renders (~6s warm). Caveat: text garble-prone on >=2-word phrases, NOT sub-second.","price_usd":"0.01","price_micro_usdc":10000,"cost_measured":{"gpu_min_per_image":0.0338,"wall_s":2.03,"vram_gb":12.18,"note":"measured 2026-10-27, tmnt:3090 resident, real PNG ~500KB"},"vram_measured_gb":null},"qwen3.5":{"id":"qwen3.5","pool":"local:8080","label":"Qwen3.5 chat (private structured inference)","description":"Qwen3.5 chat — private structured inference, OpenAI-compatible chat/completions. Billed FLAT PER CALL ($0.003), not per token, since 2026-10-28.","price_type":"per_call","price_usd":"0.003","price_micro_usdc":3000,"cost_measured":{"gpu_min_per_call":0.0143,"wall_s":0.86,"vram_gb":21.45,"tokens_measured":"23 in / 64 out","note":"measured 2026-10-27, tmnt:3090 ollama qwen3.5:9b (14B Q4 pending)"}},"qwen-image-edit-2511":{"id":"qwen-image-edit-2511","pool":"halo:8002","label":"Qwen-Image-Edit-2511 (reference-conditioned image edit)","description":"Qwen-Image-Edit-2511 — reference-conditioned image edit. Send an image + an instruction and it edits THAT image while preserving identity/composition. 8-step, denoise 0.90, ~60-90s GPU warm. The true differentiator vs plain t2i.","price_usd":"0.02","price_micro_usdc":20000,"cost_measured":null,"vram_measured_gb":null},"flux3-klein":{"id":"flux3-klein","pool":"halo:8002","label":"FLUX.3 klein (premium house-quality t2i)","description":"FLUX.3 klein text-to-image — premium, house-quality renders on Halo :8002 (ComfyUI v0.41.1, flux3 clip type). Single-pass t2i, 1024px. The premium counterpart to the fast $0.01 klein lane.","price_usd":"0.03","price_micro_usdc":30000,"cost_measured":{"gpu_min_per_image":0.1341,"wall_s":8.03,"vram_gb":null,"note":"measured 2026-10-27, Halo :8000, 1024x1024 warm, real PNG"},"vram_measured_gb":null}}}