Write, analyse, code
- Long-context chat and document analysis
- Code generation and review
- Structured output and function calling
The hard part
Models don't agree on anything.
Every provider ships its own request schema, its own parameter names, its own failure modes and its own billing unit. That mismatch is the reason "just use them all" is a backend problem, not a checkbox. It is also the part we build.
01
Normalize
One request shape in. Prompts, seeds, resolutions, safety settings and streaming formats are translated per provider, so switching models never means rewriting a workflow.
02
Route
Each job is matched to a model by what it actually needs — quality, latency, cost or context length — with automatic failover when a provider degrades, rate-limits or goes dark overnight.
03
Meter
Every provider's pricing unit is converted into one token currency and charged against one balance, so a video render and a long chat draw from the same wallet.
Coverage at launch
Four lanes, one interface.
Model line-up is added and rotated continuously — the interface stays the same when it changes.
Generate and edit
- Text-to-image across multiple families
- Inpainting, restyle, background removal
- Upscaling to print resolution
Motion from a prompt
- Text-to-video and image-to-video
- Shot extension and re-timing
- Queue with progress you can leave running
Voice and sound
- Speech synthesis and cloning
- Transcription with speaker labels
- Music and effect beds
Billing
Pay for what you generate.
No stacked monthly plans you forget to cancel. Tokens sit in one balance and are spent wherever the work happens — a paragraph, a poster or a six-second clip.
- One balance, every laneTop up once and spend across text, image, video and audio.
- Costs before you runEvery job shows its token price up front — no end-of-month surprises.
- Tokens don't expire monthlyUnused balance carries over. Work in bursts if that's how you work.
Balance
this week
128,400tokens
text41%
image29%
video24%
audio6%
Between here and launch
What happens next.
August 2026
in progress
Closed alpha
Orchestration layer and token metering running against the first model set, with a small internal group generating every day.
September 2026
next
Private beta
Waitlist invitations go out in batches, oldest signups first, with starting balances to test real workloads.
October 2026
launch
Public launch
Open sign-up, full model line-up across four lanes, and founding-member bonuses credited to every waitlist account.
Waitlist
Be there on day one.
Invitations go out in order. Join now and you're near the front when the beta opens in September.