One key per project
Issue as many keys as you need — each bound to one project, spending only that project's budget — so revoking one ends its access everywhere, with nothing left behind in personal accounts.
Point one project key at OpenAI-, Anthropic-, and Gemini-style routes. Every request is logged, scoped, and billed — so spend and access stay fully visible.
USD 5 min top-up · Credits never expire · OpenAI-compatible
OpenAI, Anthropic, Google, DeepSeek, Qwen (Alibaba), Moonshot (Kimi), Z.ai, Alibaba Cloud, ByteDance (Seedance), Kuaishou (Kling), Fish Audio
Keys, billing, attribution and access — one structure across every provider, not another vendor contract.
Issue as many keys as you need — each bound to one project, spending only that project's budget — so revoking one ends its access everywhere, with nothing left behind in personal accounts.
Every provider draws from the same credit wallet and rolls up into a single monthly bill in one currency — no more stitching together separate invoices, cycles and exchange rates before finance can close the month.
Each request is logged with its model, key, tokens and credits, and lands on a project — so the number in front of you already says which team spent it, instead of a total you have to trace backwards.
Organization → workspace → project narrows access layer by layer, with model allowlists set on the project — one place answers who can call what, instead of logging into each provider to find out.
A three-tier hierarchy — organization, workspace, project — scopes every key and every model. Access follows structure, not individuals.
Organizations, workspaces, and projects each hold their own members, roles, and credit allocations.
Each project can only call the model IDs enabled for it — enforced before traffic reaches any provider.
Every key belongs to exactly one project, with its own logs and credit usage — revocable at any time.
Click a tier (demo)
Demo: project allowlist is enforced before traffic hits providers.
Every frame below came out of the models on this platform, from the prompt shown with it. Open any one to run the same prompt yourself.
wan-2-7-t2vPrompt: A large sheet of soft sand-coloured linen settles slowly over an unseen rounded form, folds forming and rippling outward as it comes to rest. Seamless warm-grey studio backdrop. Medium shot, single movement: slow dolly push-in as the fabric settles, eye level, 50mm lens, shallow depth of field. Raking side light from the left to carve the folds into relief, warm neutral grade, subtle 35mm film grain, 24fps film look. Negative: blurry, warped text, oversaturated, watermark.
seedance-2-0Prompt: Chilled orange juice pours in a steady stream into a tall clear glass on a white stone counter, condensation beading on the glass, a few ice cubes settling. Calm steady energy. Opens on: the first stream hitting the glass base. Ends on: the glass nearly full with foam settling at the rim. Bright airy kitchen background, softly out of focus. Close-up, single movement: locked-off static, slightly below rim height, 50mm lens, shallow depth of field. Backlight through the juice making it glow, clean high-key commercial grade, 24fps film look. Negative: blurry, warped text, oversaturated, watermark.

nano-banana-2Prompt: An abstract commercial backdrop: smooth coral-to-amber gradient sweeping diagonally, with three soft-edged translucent geometric panes drifting across it and a subtle grain overlay. Deliberately empty on the left half so headline text can be placed there later. Wide shot, flat graphic composition, no perspective, even illumination with no hotspots, saturated but not neon, clean modern advertising grade, high detail. No lettering, no numbers, no logos anywhere in frame.

gpt-image-2Prompt: A clear glass latte with distinct milk layering sits on a dark walnut counter, thin steam curling from the surface, a linen napkin and two coffee beans beside it. Background of a blurred cafe interior with warm pendant lights. Medium close-up, slightly below eye level to emphasise the layers, 50mm lens, shallow depth of field. Backlit from a window behind so the steam catches the light, warm tungsten accents against cool ambient shade, rich contrast commercial grade, extreme detail, photorealistic.

nano-banana-proPrompt: A Taiwanese woman in her late 20s, straight shoulder-length black hair tucked behind one ear, calm closed-lip smile, slim build, wearing a cream linen shirt and wide beige trousers, holds an unbranded natural-canvas tote bag at hip height, turning slightly toward camera. Plain warm-grey studio backdrop, no props. Three-quarter body shot, eye level, 85mm lens, shallow depth of field. Large softbox front-left with a subtle rim light behind her shoulder, warm neutral commercial grade, extreme detail, photorealistic.
happyhorse-1.1-t2vPrompt: A Korean woman, late 20s, sleek straight black hair, poised confident expression, slender build, wearing a crisp white blazer, turns gracefully toward camera while holding up a matte black skincare bottle, tilting it slightly to catch the light. Seamless white studio backdrop, clean minimalist set. Medium shot, single movement: slow 180-degree orbit around her, eye level, 50mm lens, shallow depth of field. Bright even softbox lighting, clean high-key commercial grade, 24fps film look. Negative: blurry, distorted faces, extra fingers, warped text, oversaturated, watermark.
kling-o3-proPrompt: A Korean man, late 20s, sharp jaw, short side-parted black hair, warm brown eyes, calm smile, lean build, charcoal wool overcoat, open-collar white shirt and a Korean woman, mid-20s, wavy shoulder-length black hair, dark eyes, shy warm smile, petite build, red wool trench coat with a matching belt, small gold hoop earrings stop mid-crowd at Shibuya Crossing at night, face each other, clasp hands, and sway into a slow two-step as pedestrians blur past around them. Unbranded billboards glow red, blue, white behind them, no readable text; rain-slicked crosswalk reflects neon. Medium shot, single movement: slow dolly push-in, eye level, 35mm lens, shallow depth of field, crowd and billboards in soft bokeh. Neon rim light on wet pavement, warm romantic cinematic grade, subtle haze, 24fps film look. Negative: blurry, distorted faces, extra fingers, warped text, oversaturated, watermark.
kling-o3-standardPrompt: A Nigerian man, early 20s, short fade haircut, wide excited grin, athletic build, wearing a bright orange graphic t-shirt, throws both hands up in excitement, spinning slightly toward the camera with a burst of energy, mouth open mid-laugh. Colorful blurred bedroom background with string lights and posters. Medium shot, single movement: quick subtle handheld, eye level, 24mm lens, selfie perspective, shallow depth of field. Vibrant saturated ring-light glow, punchy energetic grade, 24fps film look. Negative: blurry, distorted faces, extra fingers, warped text, oversaturated, watermark.
Usage across every provider is metered per request and rolls up into one billing view — one billing point, credits allocated down the hierarchy.
Each request logs its model, key, token usage, and credit cost — attributable to a project and a team.
Credits flow from the organization down to workspaces and projects, so budgets are set before spend happens.
Spend across providers consolidates into one billing record — finance deals with a single counterpart.
Click a role — the evidence panel swaps. Same governance layer, three different mornings.
Keep your OpenAI SDK. Scope stays on the project key — no new auth dance.
Text, image, video, audio, and embeddings share the same governance and billing — and the platform keeps extending into enterprise AI infrastructure.
Every modality under the same governance layer.
Orchestrate multi-model agent workflows on top of the same permissions and billing you already govern.
Register and govern the tools agents can use, with access inherited from the same hierarchy.
Install the ATP skills once and your coding agent already knows the auth, the model ids, the payloads and the async media flow — no provider SDKs to explain.
Run it in your project
Skills are Markdown only — they never carry your key.
Or hand this to the agent
Install the ATP Token agent skills by running: curl -fsSL https://atptoken.ai/skills/install.sh | sh -s -- claude Then read the skills and call the ATP gateway at https://api.atptoken.ai/v1 using the atp- key in my environment.
Paste it into Claude Code, Codex or any agent that reads SKILL.md.
Skills referenceGovernance doesn't change how engineers write code — keep your SDK, point it at the ATP gateway, and access control and billing apply automatically.
Start free with a project key, or talk to us about enterprise rollout, scaled procurement, and one-on-one technical support.