The release centers on spec-accurate artifacts (Goals, Context Packets, Plans, Capabilities, Tasks, Tools, Policy, Ledger) and provides a high-level façade to run ACM-compliant workflows with one call.
This is the first public release. Where we discuss fairness and enterprise readiness, we mean enablement and design intent. Guarantees depend on the policies, datasets, and orchestrators you configure around ACM.
GitHub — ACM Framework Node v0.5.0 - https://github.com/ddse-foundation/acm/releases/tag/v0.5.0
Specificatoin: https://doi.org/10.5281/zenodo.17278997
Whitepaper: https://doi.org/10.5281/zenodo.17279075
Highlights
Contracts & Planning
Typed contracts for Goals, Capabilities, Tasks, Tools, Policy hooks, Verification adapters, and Ledger events. Structured planner produces plan candidates with rationale, tool-call envelopes, and prompt digests; deterministic selectors are available for auditability.
Runtime & Replay
Deterministic-style runtime executes validated plans with guard evaluation, retries/backoff, policy & verification hooks, checkpointing, and append-only decision ledger. Replay bundles export a single audit-ready artifact combining planner outputs, ledger entries, policy outcomes, checkpoints, and task I/O.
Interoperability & Tooling
- Adapters for LangGraph and Microsoft Agent Framework preserve ACM contracts while running on external engines.
- MCP integration (Model Context Protocol) treats external tools as first-class, discoverable capabilities.
- @ddse/acm-framework façade wires planner, runtime, nucleus config, context providers, and adapters for one-call execution.
- Examples & AI Coder demonstrate planning→execution→replay, budgeting, and streaming UX in real workflows.
Quick Start
git clone https://github.com/ddse-foundation/acm.git
cd acm/framework/node
pnpm install
pnpm build
#AgenticAI #AIEngineering #Governance #OpenSource #ACM #LLM #MCP #LangGraph #MAF #Observability #Replayability #EnterpriseAI