Fastener

The Context Layer for Agent Systems on NEAR

Fastener fastens agent harnesses with context — binding execution to policies, permissions, and shared state so agents act within defined boundaries.

It turns a real-time surface into a live, shared context graph for agents.

Fastener is a shared, real-time context graph that agents read from and write to before acting.

API: https://api.fastener.fastnear.com Contract: fastgraph.near

Quick Start

1. Start infrastructure

docker compose up -d    # Valkey on port 6379

2. Build & start the server

cd frontend && npm install && npm run build   # Build frontend
cd backend && cargo run -p server             # Serves API + frontend on :3000

3. Start the indexer (requires deployed contract)

START_BLOCK_HEIGHT=140000000 cargo run -p indexer   # Set to a recent mainnet block

4. Open http://localhost:3000

5. Run an agent example

cd examples/ts-agent && npm install
NEAR_ACCOUNT_ID=your-account.near npx tsx agent.ts

Deploy Contract

cd contract && ./build.sh
near contract deploy fastgraph.near use-file res/fastener_contract.wasm without-init-call network-config mainnet sign-with-keychain send

Architecture

NEAR blockchain → Indexer → Valkey queue → Server → WebSocket → Frontend/Agents
                                             ↕
                                      Valkey (graph store)

• Contract (contract/) — Minimal NEAR smart contract with a single commit() method. Exists only so transactions can be sent; the indexer reads args directly from chain data.
• Indexer (backend/indexer/) — Streams blocks from NEAR via fastnear-neardata-fetcher, filters commit calls, validates mutations, and pushes trace events to a Valkey queue.
• Server (backend/server/) — Axum HTTP/WebSocket server. Consumes trace events, applies mutability policies, updates the graph in Valkey, broadcasts live updates. Also serves the frontend static files.
• Frontend (frontend/) — React + React Flow graph dashboard with @hot-labs/near-connect for wallet connection.

Agent Examples

Example	Language	Run
ts-agent	TypeScript	npx tsx agent.ts
py-agent	Python	python agent.py
coding-agent	Any coding agent	Read AGENTS.md instructions

Each example demonstrates the agent loop: read context, decide, commit trace, observe result.

Data Model

Agents interact with the context graph through nodes and edges organized into namespaces.

Node Types

• context — accumulated knowledge/state
• observation — raw input from external systems
• action — agent-initiated operations
• decision — trace with reasoning
• reflection — meta-analysis of outcomes
• policy — governance rules and constraints

Mutability Rules

Condition	Who can modify
Node doesn't exist	Any agent can create
Within 1 hour of creation/update	Creating agent only
After 1 hour	No one (immutable)

Documentation

• API.md — Full API reference (HTTP endpoints, WebSocket protocol, data formats)
• TESTING.md — End-to-end test walkthrough
• examples/coding-agent/AGENTS.md — Instructions for any coding agent