BetaAgentbeam watches AI agent activity for a living, so how it handles the data it collects matters as much as what it catches. Local-first by design, redacted before it's written, and honest about what it isn't — yet.
Activity is recorded to local NDJSON files in a directory with restricted (mode 0700) permissions on the machine Agentbeam runs on — not a database somewhere else.
Known credential formats — API keys, tokens, private keys — are redacted before anything touches disk.
The collector retains the latest 10,000 events and 500 scan reports. It's a single-process local prototype, not fleet storage.
No background OS surveillance, no automatic skill installation. Sentinel is an optional Chrome extension for supported AI chat sites. Uninstrumented tools are invisible to the collector.
Exports carry SHA-256 hashes. That establishes internal consistency — the record wasn’t altered after export — not authenticity or completeness, and not a compliance certification. An event log is evidence an auditor can review, not a pass. See AI compliance and regulation for agent activity for the fuller version of this.
Found a vulnerability? Email [email protected]. We ask that you not publicly disclose an issue before we’ve had a chance to address it.
No, by default. Agentbeam is local-first — activity is recorded to local NDJSON files on the machine it runs on, in a directory with restricted (mode 0700) permissions. Nothing is sent off-device unless you explicitly configure an export.
No. Agentbeam is an early, local-first prototype — it is not SOC 2 certified and does not claim to be. Export hashes establish internal consistency of a record, not a compliance certification.
Email [email protected]. We ask that you not publicly disclose a vulnerability before we've had a chance to address it.
Yes. Known credential formats (API keys, tokens, private keys) are redacted before anything is written to disk.