Compare › Traceseal
RunVouch vs Traceseal
Traceseal (Show HN, August 2026) signs a receipt for every agent invocation: what code ran, who published it, input and output hashes, sandbox profile, all verifiable offline with one command. RunVouch proves a scheduled run happened and was done, and pages you when it was not. Both are proof; they answer different questions.
| Traceseal | RunVouch | |
|---|---|---|
| What is proven | which signed skill ran, on which inputs, in which sandbox (per invocation) | that the scheduled run started, ended, passed its evidence check and what it cost (per run, chained per day) |
| Anchor | ed25519 signatures plus a transparency log | SHA-256 chain per day, anchored in Bitcoin via OpenTimestamps |
| Verify without trusting the operator | yes, pip install traceseal-verify | yes, the chain and every day file are public, verify with the free ots client |
| Missed-run / dead man's switch | no | yes |
| Alert when the run is late, fails, loops or overspends | no | yes: Telegram, Slack, email, PagerDuty, webhook |
| Hard cost cap + pause | no | yes |
| Requires a sandbox around the agent | yes, bwrap kernel namespaces | no, wrap the command or two HTTP calls |
| Regulatory angle | EU AI Act Article 50 receipts | audit trail of unattended runs; no compliance claims |
| Pricing | not published on the site (September 2026); open source verifier | free 3 agents; $9 / $29 |
When to use which
If you have to prove to a third party which code an agent executed and on which inputs, Traceseal is built for exactly that and RunVouch is not. If you have to know at 07:00 that last night's agent ran, did the work and stayed under budget, and want a proof of that record you can hand to anyone, that is RunVouch. Running both is coherent: Traceseal seals the invocation, RunVouch watches the schedule.