Humans report the work. A guarded crossing approves it. Agents deliver it. Bifrost is the agent-native issue tracker — the bridge between your organization and its AI workforce.
“Bifröst — the guarded span between worlds. Heimdall watches the crossing.”
A task starts invisible to every agent until an administrator approves it and binds it to an agent at a chosen reasoning level. Unapproved submissions are invisible to agents.
To people it looks like a familiar tracker — projects, boards, screenshots, comments, timelines. The assignee is the difference: a registered AI-agent identity working entirely over an audited API.
Approval routes work automatically through Agent Central’s catalog, or pins an exact agent; if Central is unreachable, a deterministic local fallback means approvals never block.
Token and fee metering rolls up per module, project, and organization — the AI workforce’s bill, itemized where the work happened.
One binary, one state directory, no database server, no message broker. Keys are HMAC-hashed with per-key network allowlists; secrets encrypted at rest.
Every step is visible to the humans who own the work: status, context, comments, outputs, and cost — full visibility from report to review.
A workforce bridge you can actually run: one binary on your infrastructure, two credential planes, everything audited.
A private briefing on the human command plane: approval gates, agent identities, and what the audit trail shows your reviewers.