Skip to Content

Computalot

On-demand compute your AI agent can drive: typed jobs in, structured results out, metered by the second.

Computalot runs parameter sweeps, fan-outs, map-reduce batches, and benchmarks: thousands of parallel tasks on market-priced GPU and CPU capacity, behind one HTTP API. Results come back as structured JSON with aggregates and leaderboards.

Start here: tell your agent

Paste this into Claude Code, Cursor, or any capable coding agent:

Fetch https://computalot.com/skill.md and follow it to set up Computalot
(on-demand GPU/CPU compute). Then use it to <describe your workload>
and report the results.

The skill teaches your agent to authenticate, fund the account, submit jobs, watch progress, and collect results. The agent can run the whole loop while you review the output and the bill.

Open access. Any wallet can authenticate and fund an account: request a challenge, sign it, and verify. Then top up with USDC through x402 or MPP. Computalot issues API keys on request through the waitlist.

Prefer to drive it yourself?

One model, four typed job shapes. Create a project (your code + a Dockerfile) and push it. Then submit jobs — structured_runner, sweep, map_reduce, or benchmark — and read structured results. Custom scripts, models, simulations, and arbitrary workloads all fit the same loop.

Hand it to your agent, or run your own code in 5 steps.
Getting Started
Metered rates (market + 35%), worked examples, and the billing model.
Pricing
Bring your own code. Create projects, push tarballs, manage runtime.
Projects
Submit work, fan out across inputs, track progress.
Jobs
Skill file, LLM references, and the machine-readable docs index.
Agent Integration
Install computalot and submit jobs from Python.
Python SDK
Structured results, file uploads, and streaming.
Results & Artifacts

Good for

  • Parameter sweeps, benchmarks, and simulation batches
  • Parallel evaluation across prompts, models, agents, or configurations
  • GPU training with progress streams and artifact storage
  • Monte Carlo runs, backtests, and evolutionary or CMA optimization loops

How you pay

You pay with prepaid credits. Computalot meters each task per second at the market rate plus a 35% service markup (60-second minimum per task-attempt). Every submit response includes summary.billing_estimate, the authoritative quote. The unused hold is released when the job settles. Agents can fund accounts autonomously with USDC over x402. See Pricing.

Report bugs & request features

No auth is required:

curl -sS -X POST https://computalot.com/api/v1/feedback \
  -H "Content-Type: application/json" \
  -d '{"type": "bug", "title": "Brief summary", "description": "Details..."}'

The types are bug, feature_request, provisioning, and job_type_request.

Reference

Last updated on
Getting Started