Implemented ≠ planned
A roadmap with boundaries.
Secure self-hosted control comes first, universal gateway coverage supports it, and governed agent infrastructure expands within an explicit scope.
Updated · Public beta
Implemented in the pinned register
Verified register scope- OpenAI-compatible chat completions
- OpenAI-compatible embeddings
- Routing & fallbacks
- Weighted / load-balanced routing
- Conditional routing (rule/expression based)
- API key management
- Budgets and quotas
- Built-in guardrails
- OTLP telemetry export
Only capabilities marked Supported in the OmniSwitch register appear here. Source, version, and date matter; use the full comparison for partial capabilities.
Near term / unreleased v0.2.0
In progress- Complete and verify the v0.2.0 release artifacts and installation paths.
- Tool-capable streaming Responses/Messages subsets, OpenAI image/audio endpoints.
- Token and spend policies, extended pricing, migrations, retention, and Postgres deployment.
- Per-tool MCP discovery filtering and configurable beta token exchange.
- The redesigned authenticated operator console and corrected capability register.
This is a release work list, not evidence that a download or a published measurement exists.
Medium term / beyond v0.2.0
Planned- Dashboard OIDC SSO and an admin audit log.
- MCP OAuth protected-resource discovery (RFC 9728).
- A database-backed policy management API.
Longer term / evaluate with real workloads
Deferred- Embedding-based similarity cache.
- Downstream A2A agent proxy, persistent task orchestration, and push notifications.
- Expanded eval experiments and workload-driven API coverage.
Current A2A (Beta): SendMessage, SendMessageStream (SSE task events), GetTask, best-effort CancelTask; tasks in memory; no push notifications; no downstream agent proxy.
What this roadmap does not promise
- No dates are promised for deferred work.
- No performance ranking before a published benchmark campaign.
- No mTLS, SAML, SCIM, automatic configuration import, or unlimited API parity claim.