Cost Estimation
Understand the infrastructure costs of running ClawRun.
These are rough estimates based on published pricing as of March 2026. Actual costs depend on your usage patterns, provider plan, and LLM choice. Always check your provider's current pricing page for the latest numbers.
Cost components
Running a ClawRun instance involves four cost areas:
| Component | What it is | Typical cost |
|---|---|---|
| Sandbox compute | The sandbox where your agent runs | Varies by usage (free on Hobby) |
| Serverless functions | The orchestrator (webhooks, heartbeat, chat API) | Minimal |
| State store | Redis for lifecycle state (locks, wake times) | Free for light usage |
| LLM API | Token costs from your chosen provider | Varies by provider and model |
ClawRun itself is free and open source. You only pay for the infrastructure it runs on and the LLM API calls your agent makes.
Vercel pricing breakdown
Sandbox
ClawRun creates sandboxes with 2 vCPUs and 4 GB memory by default. Sandboxes automatically sleep when idle (default: 10 minutes of inactivity), so you're not paying for a 24/7 VM.
| Metric | Hobby (free) | Pro ($20/month) |
|---|---|---|
| Active CPU | 5 hours/month included | $0.128/hour |
| Provisioned memory | 420 GB-hours/month included | $0.0212/GB-hour |
| Snapshot storage | Included | $0.08/GB-month |
| Max session length | 45 minutes per wake | 5 hours per wake |
Max session length is the longest a sandbox can run continuously in a single wake. There is no limit on total monthly runtime. When a session ends, the sandbox snapshots and sleeps. The next message or scheduled task wakes it again and the clock resets.
Active CPU is only charged when the CPU is actually computing. Time spent waiting for network, database, or LLM responses does not count. A typical agent session has very low CPU utilization since most time is spent waiting for I/O.
Provisioned memory is charged for the entire time the sandbox is running, regardless of CPU activity.
Serverless functions (the orchestrator)
The Next.js orchestrator runs as serverless functions. These handle webhooks, heartbeats, chat streaming, and API routes. Usage is minimal since each invocation is short-lived.
| Metric | Hobby (free) | Pro ($20/month) |
|---|---|---|
| Invocations | 1M/month included | $0.60/1M |
| Active CPU | 4 hours/month included | $0.128/hour |
State store (Redis)
ClawRun uses Redis (Upstash) for sandbox lifecycle state: creation locks, next wake times, and webhook registrations. Usage is very light — a few commands per heartbeat cycle.
| Metric | Free tier | Pay-as-you-go |
|---|---|---|
| Commands | 500K/month | $0.20 per 100K |
| Storage | 256 MB | Included |
A single ClawRun instance typically uses fewer than 10K commands per month, well within the free tier.
Example scenarios
Casual use (Hobby plan)
A personal agent used a few times a day. The sandbox wakes when you send a message and sleeps after 10 minutes of inactivity. While awake, it handles any number of messages.
| Estimate | |
|---|---|
| Sandbox sessions | ~3 wakes/day, ~10 min each |
| Active CPU | ~0.75 hours/month |
| Provisioned memory | ~60 GB-hours/month |
| Redis | ~5K commands/month |
| Infrastructure cost | $0 (within Hobby free tier) |
| LLM cost | See LLM costs |
Regular use (Pro plan)
An agent connected to Telegram or Discord. The sandbox wakes on each incoming message and stays alive for longer sessions as conversations continue.
| Estimate | |
|---|---|
| Sandbox sessions | ~5 wakes/day, ~20 min each |
| Active CPU | ~5 hours/month → $0.64 |
| Provisioned memory | ~200 GB-hours/month → $4.24 |
| Snapshots | 1.5 GB → $0.12 |
| Serverless + Redis | ~$1 |
| Infrastructure cost | ~$6/month (covered by $20 Pro credit) |
| LLM cost | See LLM costs |
Heavy use (Pro plan)
An agent with multiple active channels keeping the sandbox awake for extended periods throughout the day.
| Estimate | |
|---|---|
| Sandbox uptime | ~8 hours/day |
| Active CPU | ~36 hours/month → $4.61 |
| Provisioned memory | ~960 GB-hours/month → $20.35 |
| Snapshots + serverless + Redis | ~$2 |
| Infrastructure cost | ~$27/month ($20 credit + ~$7 overage) |
| LLM cost | See LLM costs |
LLM costs
LLM API costs are paid directly to your chosen provider and are not included in the estimates above. Costs vary widely depending on the provider, model, and how much your agent uses them. Check your provider's pricing page for current rates.
ClawRun includes built-in cost tracking and budget enforcement so you can set daily and monthly limits to stay in control.
Tips for reducing costs
- Let sandboxes sleep. The default 10-minute idle timeout is a good balance. Don't increase
activeDurationunless you have a specific reason. - Use cost limits. Configure daily and monthly spending caps during deploy to prevent runaway LLM costs.
- Right-size vCPUs. The default 2 vCPUs is sufficient for most agents. Only increase if your agent runs CPU-intensive tools.
- Choose your model wisely. Smaller, faster models (Groq, DeepSeek) cost a fraction of frontier models for many tasks.