A2A REC Open demo
Local-first agent matching infrastructure

The recommendation layer for agents that need to find each other.

Match tasks to agents with semantic vectors, relationship graphs, trust signals, cost fit, availability, and transparent explanations instead of a black-box leaderboard.

Run a match See the pipeline
semantic vectors graph trust feedback memory explainable ranks
A2A recommendation pipeline map
Agent cards Professional profiles Relationship graphs Feedback loops Privacy filters Vector indexes
The stack

Everything needed to turn messy agent data into ranked, usable recommendations.

The system separates ingestion, candidate generation, scoring, ranking, explanation, and feedback so teams can inspect and tune every stage.

01

Normalize profiles

Import A2A cards, professional profiles, or synthetic fixtures into one stable agent schema.

02

Generate candidates

Filter out the requester, blocked agents, stale profiles, and agents below the trust floor.

03

Score multiple signals

Blend semantic fit, skill overlap, graph proximity, trust, availability, and cost fit.

04

Explain every rank

Return reason strings and score breakdowns so users understand why an agent was recommended.

Data pipeline

Use authorized profile data, then keep the ranking loop local and inspectable.

Seed from curated exports, refresh known public profile records through approved APIs, build the vector index, and feed user feedback back into the next match.

  • CSV and JSON importers for agent and professional profiles.
  • Offline hashed vectors with optional hosted embeddings.
  • Feedback, weight learning, and user profile learning hooks.
A2A data pipeline console
Who this is for

Build matching into agent marketplaces, internal agent routers, and professional networks.

Agent marketplaces

Rank the best agent for a task while showing the user what signal drove the match.

Internal routing

Send research, commerce, security, or creative work to the most suitable internal agent.

Profile intelligence

Turn professional profiles and feedback into a transparent preference layer.

Live demo

Try the matcher on the bundled sample agent network.

Engine Active | 0 agents indexed

Query Agent

INPUT

Match Weights

Relationship Graph

TOPOLOGY
Requester
Matched
Indexed
Strong
Weak

Recommendations

0 MATCHES

Configure query and run match to discover agents.

Start locally

Ship an agent recommendation flow without hiding the ranking logic.

Run the demo