Roadmap
What's shipped, what's being built, and what's coming next. Updated with each release. Follow GitHub Releases for shipping notifications.
Shipped
14 itemsReleased and stable in the latest version.
createGovernance(), enforce(), register(). 13 condition types. Boolean combinators. Priority-ordered evaluation.
createGovernanceMiddleware() and createGovernedMCPProcessor() — wrap every tool call with a single line.
createGovernedTools() wraps any AI SDK tool definition with before-action enforcement.
detectInjection() across 7 categories: instruction override, role switch, data exfil, command injection, goal hijacking, prompt leakage, and more.
kill(), killAll(), revive(), reviveAll() at priority 999. Per-process in OSS; distributed in Cloud.
Composite 0–100 score across auth, guardrails, observability, audit, oversight, compliance, identity. L0–L4 levels.
governTool() wraps any DynamicTool or StructuredTool with full policy enforcement.
createIntegrityChain() with append(), verify(), and export(). Tamper detection with exact broken-link location.
assessCompliance() mapping to Articles 9, 11, 12, 14, 15, 50. getDaysUntilDeadline().
Auto-migration on first connect. No migrations file needed. Zero config beyond connection string.
on('enforce', 'register', 'kill', 'score') lifecycle hooks. Latency histograms, block ratios, enforcement rates.
governAgent() for @openai/agents. Govern an entire OpenAI agent with one call.
createEnterpriseGovernance(): RBAC, multi-tenant namespacing, fleet analytics, policy templates (fintech, healthcare, saas).
Broad coverage across the enforcement hot path, audit-chain verification, and framework adapters. Scope-honesty pass in v0.11 trimmed 5 unused modules; test suite still passes clean.
In Progress
3 itemsActively being built. Expected in Q2 2026.
Middleware adapters for Apollo Server, tRPC, and Hono that govern AI-generated query execution before resolver runs.
Detect when a registered agent's tool usage pattern drifts from its baseline. Alert or auto-kill on statistical anomaly.
Native adapter for Model Context Protocol servers. governance-sdk/plugins/mcp ships today; roadmap item covers response-side sanitization and credential redaction.
Planned
3 itemsOn the roadmap with clear scope. Not yet started.
Compile the enforcement hot path to WebAssembly so governance runs natively in Cloudflare Workers, Deno Deploy, and other Wasm-friendly runtimes. Latency targets TBD after prototype.
Hosted version of the admin dashboard with live fleet metrics, audit log search, compliance reporting, and team management.
governance validate, governance apply, governance diff — manage policies from git, deploy via CI/CD pipeline.
Exploring
3 itemsResearch phase. No commitment on timeline.
When agent A calls agent B, propagate governance context and enforce trust-level requirements across the call graph.
Use Z3/SMT solvers to prove that a policy set can't have conflicting rules that produce undefined outcomes.
Conflict-free replicated policy updates across distributed agent instances with guaranteed consistency.
Open an issue or start a discussion on GitHub Discussions. We read everything.