Flat-rate inference for open-weight models. Use it from Cursor, VS Code, OpenCode, Hermes Agent or any harness: you bring the tool, we bring the engine. One fixed monthly subscription.
How fast the model writes, letter by letter.
Your ideas flow without cuts
The speed adding every user at once.
Scales with your team, no degradation
How long until the first word shows up.
Answers before you finish the thought
The model reuses context it already read.
Long sessions, constant speed
How long it ran without a single incident.
Weeks of real work, that stable
TU HERRAMIENTA
CAPLESS GATEWAY
MOTOR DE INFERENCIA
GPU DEDICADA
What runs beneath every prompt.
Dedicated high-performance GPUs
Whole hardware for you: shared only with whoever you decide.
No noisy neighbors, no throttling
Nobody else eats your capacity when you need it most.
24/7 health checks & auto-recovery
If something breaks, it stands back up before you notice.
Zero. Ephemeral RAM only
What you write never touches a disk. Ever.
No catalog bloated with 500 redundant models. Only weights proven at code generation, refactoring and terminal utilities.
~3B-active Mixture-of-Experts for TypeScript, Python, Go, Rust and SQL generation. Outstanding context tracking on agent instructions and tool calling.
A lightweight model tuned for fast reasoning, commit generation, PR summaries and log filtering — without burning your heavy-compute window.
High-dimension embedding models for local RAG over your code repo, plus Whisper STT and Kokoro synthesis bundled into the plan at no extra cost.
Not ours, not anyone's. Your prompts and completions exist only in GPU RAM during inference and are purged when it ends. Zero-logs, zero-training.
Structured as terminal receipts. Zero credit top-ups in the middle of a night deploy.
Fair-use limits published in the FAQ. No fine print.
We looked at the real economics of coding with AI every day.
| FEATURE | TYPICAL PAY-PER-TOKEN | ENTERPRISE ($200/mo) | CAPLESS ($10 - $20/mo) |
|---|---|---|---|
| Cost predictability | None. Surprise top-ups at 3 AM when an agent iterates on its own. | Fixed, but absurdly inflated with forced annual contracts. | 100% predictable. Fixed monthly charge. Zero surprises. |
| Privacy & Logs | They keep prompts for 30 days for "abuse auditing". | Complex legal agreements with exception clauses. | Zero logs, zero training. Direct inference on ephemeral RAM. |
| Client setup | Complex keys and proprietary headers. | Requires VPN software or bloated SSO auth portals. | 1 URL in baseURL. Works instantly on Cursor, Cline and CLI. |
| Hardware transparency | Hidden behind opaque load balancers and silent throttling. | Quarterly aggregated metrics. | Public telemetry and real-time GPU status. |
Any OpenAI-compatible client works by changing one line: Cursor, VS Code, OpenCode, Hermes Agent, Aider, Continue… same pattern — baseURL + API key.
Full documentation with exact snippets at beta launch.
It's a flat rate with generous windows of real usage. We don't charge per extra token and we don't cut your service if you have a sensible work spike. The 4-hour rolling windows are sized so no normal developer ever saturates them.
Published and generous: a monthly quota per plan plus an automatic window reset every few hours, so everyone gets the same speed at any hour. If you ever hit it, your priority dips for a while — zero extra charges.
They run on our dedicated GPU instances. We operate a Zero-Logs architecture: your prompts and code snippets exist only in volatile memory (VRAM) during inference and are purged immediately. They are never written to disk or used for re-training.
Yes. Any tool that accepts an OpenAI-compatible endpoint (baseURL) works instantly by changing one config line and pasting your Capless API key.
Because they can be audited, the real cost of serving them is far lower, and that saving is exactly what makes the flat price possible. We don't offer 400B models: we host 30B MoE models that actually fit on accessible hardware.
We release access in controlled waves as we provision capacity, so generation speed never drops. Join the waitlist and we'll email you when your wave opens.
The beta opens in waves — reserve your spot today.
Reserve your spot with your developer email. You'll receive your API key when the beta opens — cost today: $0.
RECEIPT #00 · WAITLIST · COSTE HOY $0
Form handled by Tally. No spam; we only email your API key when the next GPU spins up.