Claude.ai Browser Chat Telemetry in Beacon | Asymptote
What is Claude.ai?
Claude.ai is a browser surface, not a terminal harness, but the security question is the same: what prompt, response, tool, and usage activity happened on the machine?
Collection path: Managed Chrome MV3 extension over local OTLP
Authoritative date: August 24, 2026 from Browser extension in the Beacon CLI changelog.
How Beacon reads Claude.ai
The optional browser extension reads supported Claude.ai chat streams and posts OTLP GenAI logs to the local collector on 127.0.0.1:4318. Beacon normalizes those events under the claude_web harness name so browser chat stays separate from Claude Code.
The result is normalized into Beacon's endpoint event schema so security teams can investigate activity by harness, repository, session, user, model, command, file, tool, MCP server, or approval context where the runtime exposes those fields.
Telemetry coverage
- Prompt telemetry from supported Claude.ai chat streams
- Assistant response telemetry when retained by the selected browser extension mode
- Tool call and token usage telemetry where exposed by the chat stream
- Local runtime.jsonl output and existing customer-managed forwarding paths
What you can do now
- Review Claude.ai conversations alongside local agent sessions.
- Separate browser chat from CLI activity with the
claude_webharness name. - Choose full, redacted, or metadata-only retention before enabling collection.
Get started
- Update to a Beacon release that includes browser extension support.
- Build the extension from
browser-extension/and load it unpacked in Chrome. - Confirm events land in the Beacon runtime log and your existing forwarding destination.