trust-ledger-os

Trust Ledger OS is a trust and risk control plane for AI teams. Its core idea is: every high-impact change must be reviewed, tracked, and recorded before it reaches customers or funds.

Purpose and Positioning This project combines a public landing page with four phased implementation scaffolds, and mirrors the same roadmap into reusable package boundaries so that the application, documentation, and future library work stay aligned. It is aimed at teams that need to establish auditable decision processes for AI agent actions, code changes, and spend requests.

Main Capabilities

• PRISM: Provides runtime observability for agent invocations, approvals, and latency.
• Prelint: Pre-merge product review, used to detect policy drift, bad defaults, and risky logic.
• GIDE: Provides secure offline editing and emergency fixes when the network is untrusted.
• Decision process: Capture a code change, spend request, or agent action; review it against product policy and trust rules; attach PRISM traces to runtime decisions; produce an approve, reject, or counter result; store the decision in the ledger for reuse in future runs.
• Decision substrate: CHP decision records, including the R0 gate, deterministic base scoring (guardrails 40 / bounded outcomes 30 / consistency 30), configurable per-domain floors (general 70, finance 100, blockchain/DeFi 85), substrate-enforced human lock (TRUST_LEDGER_CHP_REQUIRE_HUMAN_LOCK, enabled by default), and an append-only JSONL ledger sealed with SHA-256 body_sha256 and re-verified on every read. This substrate is a TypeScript port of the validated primitives in erp-control-plane, with canonical JSON implemented via @cubiczan/chp.

Repository Structure

• /: Landing page and roadmap
• /foundation: Phase 1 scaffold
• /framework-benchmark: Phase 2 scaffold
• /research-reasoning: Phase 3 scaffold
• /production-controls: Phase 4 scaffold
• /decisions: Decision ledger (CHP decision substrate), GET /api/decisions reads in real time
• foundation-kit/: Shared agent glossary, decision trees, model/tool matrix, and prompt skeletons.
• framework-benchmark-suite/: Framework profiles, comparison cases, and scoring criteria.
• research-reasoning-kit/: Research pipeline stages and reusable reasoning patterns.
• production-controls-kit/: Evaluation tooling, drift monitoring, guardrails, and release gate scaffolding.
• decision-substrate/: CHP decision records.

Distribution Formats

• packages/npm/: npm-friendly manifest and directory exports, published as @cubiczan/trust-ledger-os@0.1.2.
• packages/python/: PyPI-friendly manifest and directory exports, published as trust-ledger-os==0.1.2.
• packages/mcp/: MCP-ready tools and resource catalog, published as trust-ledger-os-mcp@0.1.2.

Package Commands

• npm package build: npm --prefix packages/npm run build
• MCP package build: npm --prefix packages/mcp run build
• MCP server run: npm --prefix packages/mcp start
• Python CLI: PYTHONPATH=packages/python/src python3 -m trust_ledger_os.cli --manifest
• Vercel Python entrypoint: src.trust_ledger_os.vercel_app:app (from packages/python)

Release Workflow

• GitHub Actions workflow: .github/workflows/release.yml
• Triggered by v* tags or workflow_dispatch
• Secrets required for release mode: NPM_TOKEN, PYPI_API_TOKEN
• The workflow builds the Next app, the npm and MCP packages, and the Python package before publishing.
• The 2-minute video plan and FFmpeg render commands are located in docs/TRUST_LEDGER_OS_VIDEO.md.
• Generated assetsLocated in public/demo: trust-ledger-os-demo.mp4, 01-hero-desktop.png, 02-how-it-works.png, 03-tool-stack.png, 04-builderbase-fit.png, 05-demo-deliverables.png, 06-hero-mobile.png.

Implementation Roadmap

1. Foundational kit: Shared agent glossary, model/tool matrix, and guidance on when to use agents.
2. Framework benchmarks: Compare LangGraph, OpenAI Agents SDK, AutoGen, PydanticAI, and LlamaIndex.
3. Research and reasoning: Fundamental analysis, deep search, and reusable reasoning templates.
4. Production controls: Evaluation tools, drift checks, tracing, guardrails, and approval gates.

Development Notes

• Keep shared data in package folders, not in page components.
• Import from each package's src/ entry before package exports are formalized.
• Add new phase work first as reusable types and data, then wire into routing and documentation.
• Treat README.md and docs/TRUST_LEDGER_OS.md as the single source of truth for the roadmap.

Applicable Scenarios and Limitations Suitable for teams that need auditable, reusable decision records for AI agent actions, code changes, and spend requests, with a particular focus on policy drift, risk logic, and human approval locks. The project is currently in the scaffolding and early release stage (0.1.2), and some interface screenshots may be demos or sample data. Information on fees, pricing, platform support scope, and specific outcomes is not yet available.