Skip to content

ROADMAP.md

Latest commit

 

History

History
27 lines (19 loc) · 2.39 KB

ROADMAP.md

File metadata and controls

27 lines (19 loc) · 2.39 KB
Raw

AgentOS Project Roadmap

Completed Milestones

  • Phase 1-4: Full AI gateway implementation featuring model routing, response caching, policy filtering, RBAC, multi-cluster federation, and a Kubernetes operator.
  • Phase 5: Semantic caching using embedding vector similarity, cost optimization suggestions, custom request/response text transforms, priority-queue-based load shedding, WebSocket proxying, GraphQL admin console, and WebAssembly (WASM) policy SDK.
  • Phase 6: MCP remote gateway supporting JSON-RPC routing, dynamic tool allowlisting/denylisting, operator review execution path, and queue storage.
  • Phase 7: Task-scoped credential broker enabling temporary Github App JWTs, dynamic AWS STS SigV4 credentials, Vault database credentials, and dynamic credential metadata injection into the evidence chain.
  • Phase 8: Evidence export and verification tooling CLI (agentctl verify, agentctl evidence) with three tuned policy packs for coding agents.

Enterprise-Grade Capabilities

  • Tier 1 (Core Integrity): Typed resource model checks, TaskManifest definition, manifest execution drift detection, cryptographic capability tickets, policy dry-runs and diffs, sandbox execution helpers, and formatted human-auditable evidence.
  • Tier 2 (Integrations & Compliance): Behavioral session rules, Slack Block Kit & GitHub PR approval webhooks, enterprise identity tracking, separation of duties compliance, and signed policy supply-chain keys.
  • Tier 3 (Reliability): HA setup, automated db recovery/retention strategies, threat modeling docs, and security guides.

Runtime & Framework Integrations

  • Notifications: Alert systems triggering automatically on approval states or Slack/GitHub reviews.
  • Kill Switch: Sessions automatically blocked when risk thresholds are exceeded.
  • Drift Enforcement: Option to switch from warn to enforce mode to actively block out-of-scope actions.
  • Formatted Reports: Auto-rendering Markdown and HTML reports suitable for security officers and auditors.

Adoption & Tools

  • Phase 9: Governed Coding Agent Starter Kit including 3 tuned policy packs, developer editor integrations, Docker/Helm/Terraform templates, test harness, and sample outputs.
  • Phase 10: Fully documented PR-writer scenario walkthrough, quick installation scripts, and design-partner onboarding material.