check_memoryBefore a heavy step, your agent sends free memory and the job. It gets a warning and run_local or offload.
Cloud for Botsinfrastructure for your bots
Your bot is out of memory. Give it more compute.
Before a heavy step, your agent checks free memory. When the machine it runs on is running short, the build, install, test suite or headless browser goes to a fresh cloud sandbox on your own account: Vercel Sandbox by default, or E2B, Daytona, Modal and Workers Sandbox. You bring the keys and pay your provider directly. We charge a flat fee and never resell compute.
500 calls/day free · Pro $9/mo flat · your keys, your provider bill · keys sealed AES-256-GCM
coding-bot · laptop · floor 2 GB
free memory3.1 GB
offloaded0 jobs
- npm run buildweb build · needs ~2.4 GB2.4Gqueued
- npx playwright testheadless browser · needs ~1.5 GB1.5Gqueued
- git statuslight command · needs ~20 MB20Mqueued
- pytest -n 4test suite · needs ~1.2 GB1.2Gqueued
// run_job results appear here
check_memory · waiting for the next job
Sample: Only 950 MB free, under your 2 GB floor. npm run build needs about 2.4 GB (rough guess for a web build), leaving nothing. Watch-only: would offload to Vercel Sandbox. Nothing ran.
Works with the agents you already run
- AClaude
- {}Codex
- CChatGPT
- GGrok Bot
How it works
Check memory. Offload the heavy part. Keep your own bill.
save_provider_keyOnce, from your own machine: save a Vercel, E2B, Daytona, Modal or Workers Sandbox key. It is sealed and never shown again.
run_jobYour agent sends the command. A fresh sandbox on your account runs it, returns exit code and output, and stops.
You pay your provider, not us
Flat connector fee. Compute at your provider's own price.
For solo builders: every provider below has a free tier or included credit, and you keep your own account and bill. Compute Connector never resells compute or adds a markup; it only connects your bot to the account you already have. Here is what each lists today.
Vercel SandboxHobby: 5 CPU-hrs/mo included · Pro: $0.128/CPU-hrpricing page ↗E2BHobby $0 + one-time $100 credit · Pro $150/mo + usagepricing page ↗Daytona$200 free compute, then $0.0504/vCPU-hrpricing page ↗ModalStarter $0 with $30/mo compute includedpricing page ↗
Compute Connector Pro$9/moflat fee, any provider, no markup
List prices checked Oct 17, 2026 on each vendor's pricing page. Compute Connector does not replace them.
Pricing
Watch free. Pay when it acts.
Free resets daily and is not a trial. Pro is $9 a month.
Free · watch-only
$0
Watch-only mode. 500 calls a day: check_memory warns when your machine is under its memory floor, run_job logs a would_offload instead of running, and weekly_report shows how often your bot ran short. Nothing runs and no key is stored.
- Memory warning before every heavy job
- would_offload logged instead of running
- Weekly report: how often your bot ran short
- Stores no keys, runs nothing
Pro
$9/month
Acting: run_job sends heavy jobs to a fresh sandbox on your own Vercel Sandbox, E2B, Daytona, Modal or Workers Sandbox account and returns the exit code and output. Provider keys sealed with AES-256-GCM, sandboxes stopped after each job, 50,000 calls a day. Flat fee: your provider bills compute to you directly.
- run_job on Vercel Sandbox, E2B, Daytona, Modal, Workers Sandbox
- Provider keys sealed (AES-256-GCM), never returned
- Fresh sandbox per job, stopped right after
- Flat fee, no compute markup, 50,000 calls a day
Worked example
Your coding bot runs on a laptop with 950 MB free and is about to run npm run build (about 2.4 GB). On free, check_memory says "Only 950 MB free, under your 2 GB floor" and logs would_offload, and Monday's weekly_report counts 14 jobs that would have moved. With Pro at $9/month, you save a Vercel token once, and run_job builds it in a fresh Vercel Sandbox in about 40 seconds, returns exit 0 and the last lines of output, and stops the sandbox. Vercel bills those seconds to your own account (on Hobby they come out of your 5 included CPU-hours); we never add a markup.
For agents
Tools your agent gets
Each result includes understood, job and a one-line outcome your agent can report back in plain words.
check_memoryBefore a heavy job, send how much memory the machine your agent runs on has free (free -m, os.freemem()) and what you are about to run. Get a memory warning and a verdict: run_local, or offload with the provider to use.
run_jobPro: run a shell command in a fresh cloud sandbox on your own provider account (vercel (Vercel Sandbox, default), e2b, daytona, modal, workers (Workers Sandbox)) and get the exit code and the last 16 KB of stdout and stderr back. Send small input files with it.
save_provider_keyPro, owner only: save one provider key, sealed with AES-256-GCM before it reaches the database. Providers: vercel (Vercel Sandbox, default), e2b, daytona, modal, workers (Workers Sandbox).
list_providersEvery provider this connects to, how each one runs, and (with your key) the provider keys you saved: name, provider, fingerprint, default, last used. Never shows a key.
delete_provider_keyPro, owner only: delete a saved provider key by name. Jobs on that key stop working at once.
provider_setupStep-by-step setup for one provider: where to get the key, what to pass to save_provider_key, links to its docs and price page, and for Modal or Workers Sandbox the bridge code to deploy on your own account.
weekly_reportPlain-English summary of the last 7 days: memory checks, low-memory warnings, jobs that would have been offloaded (free) or ran (Pro) by provider, minutes and failures. Read-only; sends nothing.
usage_and_upgradeUsage today + upgrade link (not metered).
Connect
One line to install
claude mcp add --transport http compute-connector https://compute.cloudforbots.com/mcp
MCP URL https://compute.cloudforbots.com/mcp
REST POST https://compute.cloudforbots.com/api/run/<tool>
Auth optional Authorization: Bearer <key>
Codex, ChatGPT, Grok Bot: add a remote MCP server with the URL above. Skill file: SKILL.md · catalog entry: tool.json
FAQ
Questions
What problem does this solve?
Agents run heavy jobs (builds, installs, test suites, headless browsers) on whatever machine they live on. When that machine is low on memory, the job freezes it or gets killed. Compute Connector tells your agent before it happens and moves the job to a cloud sandbox on your own account.
Do you sell compute?
No. You bring your own provider key and the provider bills you directly at its own price. We charge a flat $9/month connector fee and never mark up or resell compute (provider terms don't allow resale anyway).
Which providers work?
Vercel Sandbox (the default), E2B and Daytona through their public APIs. Modal and Workers Sandbox through a small bridge you deploy on your own account; provider_setup gives you the code.
How are my provider keys stored?
Sealed with AES-256-GCM before they reach the database, bound to your account and the key name. No tool ever returns a key, and keys, commands, environment values and output are never written to our logs.
What do you keep about my jobs?
Only the first word of the command (like npm), the provider, how long it ran, the exit code and memory numbers, for your weekly report. Never the full command, files, env or output. 90-day retention.
Will a sandbox keep billing me after the job?
No. Each run_job creates a fresh sandbox, runs your command and stops it right away. If your agent also opens its own long-lived sandboxes, Sandbox Timer flags the idle ones.
What is watch-only mode?
The free plan warns when memory is low and logs every job that would have been offloaded, with a weekly report. Nothing runs and no key is stored. It is not a trial.
How big can a job be?
Up to 240 seconds per run_job call, 50 input files and 500 KB sent with the job, and the last 16 KB of stdout and stderr back. Bigger inputs: clone or download them inside the job.
Can it run on my own servers?
Not yet. Today it connects to the providers above. Bring your own key is the whole idea.
What does it cost?
Free: watch-only, 500 calls a day. Pro is $9 a month, or $90 for a year, plus whatever your provider charges you. It is also included in the Cloud for Bots All Access bundle.