Agent · Browser · Tools · Models
Aexol is the full-stack development environment — agent, browser, tools, models — and it hands you the controls: any model in the catalog or your own, a dedicated GPU session from $3/h (the same session is $15+/h elsewhere). Work how you want, when you want, on what you want.
Dedicated GPU session from $3/h · Any model in the catalog or your own · No training on your data
- Agentready
Always-on sessions that edit files, run commands and work with Git.
- Browseron-call
Drive any of them from a tab — no SSH, no port forwarding.
- Toolsready
Terminal, Git, MCP servers and a real editor in one loop.
- Modelsready
Any model in the catalog, or your own open weights.
Run any model in the catalog or your own, on a dedicated GPU session from $3/h.
Your models. Your GPUs. Every machine you own.
Rent a GPU and host your own models on it — you pay by the hour, not per token. The same harness reaches every machine you own from one browser tab or a phone, and models come up from the API.
Your own models, on a rented GPU
Rent a GPU, boot the model, chat. You pay by GPU-hour — the tokens it serves are free. A cold model says so; starting it from the chat warms it to READY.
start → READY
One harness, every machine
Install Spectral on each machine and run it. Laptop, workstation and server become live agents in one browser tab — and the phone drives the same machines. No SSH, no port forwarding.
spectral serve
Stand models up from the API
Start the GPU run over the API, then call the model on an OpenAI-compatible endpoint any client can talk to.
POST /v1/chat/completions
Browser tab
agent view
Phone
same machines
relay
Laptop
live agent
Workstation
live agent
Server
live agent
The fleet behind it
FP4 = 4-bit quantized decode · total = aggregate fleet throughput · billed by GPU-hour, tokens served are free
Explore modelsWhere we're heading.
Three goals define what we're building toward — speed, access, zero friction.
1000tok/s
Output speed on a model that can agentic code, with only 10–30 B active parameters.
Private, for any team size
Even very small teams get private endpoints — at the best and lowest price per hour.
One click, out of the box
A ready private endpoint in a single click. Serverless, nothing to operate.
Made in the EU. Hosted wherever you want. No training on your data.
Made in the EU
Built by a team in the European Union.
Hosted wherever you want
Inference hosted in your region — your code never leaves it.
Run it locally
Inference on your own hardware: an open-weight model (e.g. Qwen 3.8 27B, Apache-2.0) on a GPU you own or rent.
No training on your data
We do not allow providers to train on your data.
Always-on sessions that edit files, run commands and work with Git.