Skip to content

Latest commit

 

History

History
118 lines (89 loc) · 5.28 KB

File metadata and controls

118 lines (89 loc) · 5.28 KB

AgentLedger — Roadmap

Last updated: February 2026

Vision

Build the trust and privacy layer for the agentic ecosystem — where orchestrators discover and evaluate agents based on data handling transparency, operator accountability, and per-capability trust.

We complement MCP and A2A by answering: who should I trust with my data?


Phase 1 — Trust Foundation (Current)

Core registry with ADHP, operator verification, and per-capability scoring.

Completed ✅

Feature Description
REST API (FastAPI) Register, search, update, delete agents
PostgreSQL + JSONB Persistent storage with flexible fields for future extension
ADHP fields Agents declare data handling level (open → zero-trace) with full property set
ADHP search filters Filter agents by minimum level, jurisdiction, third-party sharing policy
API key authentication Write operations require API key, reads are public
Rate limiting (Redis) Abuse prevention with graceful fallback
Keyword search Search by name, description, capabilities
Web dashboard Browse and search agents
Docker deployment Containerized with Docker Compose on shared network
Unit tests 47 tests, 84% coverage (pytest)
OpenAPI documentation Auto-generated Swagger/ReDoc
Seed data Example agents for demo
Database migrations Alembic with versioned schema changes
Batch scripts run-dev, stop-dev, run-tests, seed-db, status, view-logs
ADHP Specification v0.1 Full spec — 5 levels, delegation cascading, third-party sharing matrix
KYC Specification v0.1 Full spec — 4 tiers, GDPR-compliant, operator data model

In Progress 🔨

Feature Description
Operator system operators table linked to agents, verification tiers
Admin CLI Scripts for verification management, GDPR export/deletion
Per-capability scoring Quality/confidence scores per capability (contextual trust maps)
Privacy compliance tags Agents declare regulatory compliance: GDPR, HIPAA, CCPA, POPIA, etc.
Agent status Active/suspended/deprecated status with admin control
Public deployment Domain + HTTPS via Caddy
Actionable demo Orchestrator simulation choosing agents based on ADHP requirements

Planned 📋

Feature Description
Admin dashboard Web UI for KYC review queue, operator management, audit log
Privacy policy page Dashboard page covering GDPR rights for operators
MCP compatibility Ensure agent schema is compatible with MCP server metadata
A2A Agent Card alignment Ensure agent schema maps cleanly to A2A Agent Cards
Health checks & monitoring Uptime monitoring, alerting

Phase 2 — Ecosystem Integration

Native protocol support, federation, and enhanced discovery.

Feature Description
Federation with MCP Registry Pull server metadata from official MCP registry, add trust layer
A2A protocol support Support A2A Agent Cards for agent discovery
Semantic search Embedding-based matching of orchestrator intent to agent capabilities
OpenAPI auto-registration Paste an OpenAPI spec URL, auto-register as an agent
Usage observability Track which agents are queried, how often, by whom
Delegation chain validation Verify that sub-agents maintain required ADHP levels
Webhook notifications Notify subscribers when matching agents are registered
KYC document management Encrypted upload, admin review workflow, retention enforcement
GDPR automation Data export, deletion endpoints, retention enforcement

Phase 3 — Verified Trust at Scale

Move from self-declaration to independently verified trust.

Feature Description
Verified badge system Organizations pay for KYC + ADHP compliance verification
Automated ADHP auditing Auditor agents test registered agents with tracking markers
Reputation scoring Aggregated trust scores based on usage, audits, and peer feedback
Blast radius tracking Map agent dependency chains for risk assessment
Re-verification workflow Expiry notifications, renewal process
Third-party KYC integration Provider integration for scalable identity verification

ADHP as Standalone Standard

In parallel with the registry development, we are developing ADHP (Agent Data Handling Policy) as a standalone open specification that can be adopted independently:

  • Publish as a separate repository with Apache 2.0 license
  • Propose as an MCP Extension via the SEP process
  • Seek adoption across the agentic ecosystem
  • Engage with AAIF community

The registry is the reference implementation, but ADHP is designed to work with any registry or discovery system.


Infrastructure

Item Status
VPS (Hetzner CAX21 ARM64) ✅ Running — Ubuntu 24.04, secured
Docker + shared services ✅ Running — Caddy, PostgreSQL 16, Redis 7
GitHub repo ✅ Active — branch protection, secret scanning
CI/CD pipeline 📋 Planned
Automated backups 📋 Planned
Domain + HTTPS 📋 Planned
Monitoring & alerting 📋 Planned