A standalone, multi-tenant gateway that sits between your agents and the outside world — so they can act, without ever touching your credentials.
Pre-release
Overslash is under active development and not yet ready for production use. APIs, schemas, and behaviors will change without notice.
AI agents that touch external services — GitHub, Gmail, Stripe, Slack — hit the same wall every time, and every agent platform rebuilds it from scratch, badly:
The auth code ends up coupled to the agent loop, permissions become prompt-based ("please ask before sending"), and secrets leak into conversation context.
Overslash extracts all of that into a single service with a clean API. It is purely an auth and identity layer: it answers one question — "is this identity allowed to do this action with these credentials?" — and if the answer is yes, it makes the authenticated HTTP call.
It is deliberately not an agent framework, an LLM router, or an orchestrator. It doesn't run prompts, schedule work, manage compute, or track which agents are online. It owns identity, secrets, OAuth, permissions, approvals, execution, the service registry, and the audit trail — and nothing else.
Point your agent at Overslash, declare the services and scopes it needs, and the same backend is reachable three ways — so any HTTP client, shell-capable agent, or MCP-aware editor can use it without rebuilding the plumbing:
overslash) — for shells and scripts, and to self-host the whole product from a single binary.