OAuth tool layer
15 providers, 36 tools. Pinterest, Reddit, X, Stripe, GitHub, Slack, more. Tokens stored in encrypted vault. Agents call tools without ever seeing the secrets.
Universal control plane for fleets of AI agents. OAuth tools, audit trail, credential vault, multi-runtime adapters. Apache 2.0. Self-hostable. Public repo coming soon.
Closed-source control planes ask you to trust them with your AI agents' credentials, prompts, and audit logs. We don't think that's a reasonable ask.
OpenTac will ship under Apache 2.0 and be self-hostable. When the repo lands, you'll be able to inspect the encryption code, read the audit log schema, and vet every OAuth integration. Until then, this page describes the architecture honestly — no live screenshots faked, no metrics we can't show.
We already use OpenTac internally as the control plane for every T6AI agent fleet. The same code you'll be able to run, we run. No closed-source SaaS pivot planned.
The boring infrastructure every fleet of AI agents needs, but nobody wants to write twice.
15 providers, 36 tools. Pinterest, Reddit, X, Stripe, GitHub, Slack, more. Tokens stored in encrypted vault. Agents call tools without ever seeing the secrets.
AES-256-GCM encryption. Per-tenant key separation. Master keys never written to disk. SOC 2-style discipline without the audit theater.
Every action by every agent, timestamped and immutable. Loi 25 art. 14 accountability. Queryable via API or directly from the database.
Adapter pattern for Hermes Agent, OpenClaw, Letta, CrewAI, and any future runtime. Your agent code stays portable.
Pub/sub between agents and humans. Inbox/outbox semantics. Hop-count anti-loop. Reactions, metadata pills, action cards.
Semantic memory layer with portable file backend. Switch from PGLite to Supabase to Pinecone without changing your agent code.
GitHub repo opens to the public when V1 is hardened. One email when it lands, nothing else.