Base.com

Entity-Graph Retrieval Improves AI Agent Memory

Norax AI published a technical blog post on 2026-06-27 describing an entity-graph retrieval method to improve memory retrieval for AI agents. The approach builds an entity graph from a memory store by extracting named entities, creating co-occurrence edges, weighting edges by co-occurrence counts, and grouping entities into communities using the Louvain algorithm. At query time the system extracts entities from the query, finds their communities, computes entity overlap between query and candidate memories (with direct matches scored higher and same-community matches given a smaller boost), and boosts memories that share entities or community membership. The author argues this method finds related memories that semantic/embedding search misses, requires no additional training or external API calls, and adds minimal latency.

x402 Reaches 119M Transactions; Accountability Layer Needed

The Base blockchain’s x402 payment flow has recorded over 119 million cumulative transactions with more than $35M total volume, and nominal gas costs (< $0.0001 per transaction). Cloudflare has integrated x402 into pay‑per‑crawl tooling and nous research uses it for per‑inference billing of the hermes 4 model, demonstrating production viability beyond developer experimentation. The author argues a missing accountability layer is required when autonomous agents act as financial principals spending on behalf of humans. MnemoPay positions itself as that accountability layer — offering an SDK that adds intent-linkage and signed receipts, an agent creditworthiness score (“Agent FICO”), and provenance for authorization — without changing the underlying payment rail. MnemoPay reports 672+ tests, v1.0.0-beta.1, and ~1.4K weekly npm downloads. The post frames this layer as essential as x402 usage moves toward consumer‑scale agent commerce.

Coinbase launches AI-agent trading and payments tool

Coinbase announced "Coinbase for Agents," an agent-enabled product that lets AI agents execute trades and autonomously pay for premium research, data APIs and on-demand compute. Agents can be connected to a user’s main account or run in a separate sandbox; they can use Coinbase Advanced tools (including TradingView charts) to analyse and execute trades. The agents currently support crypto spot markets and derivatives, with planned support for equities and prediction markets and forthcoming custom limits on trade size and spending. Coinbase is using the open x402 payment protocol (built with AWS, Anthropic, Circle and Near) to enable agentic payments without requiring user logins or subscriptions. The system can interoperate with ChatGPT and Claude via Coinbase’s MCP server. Regulators and industry players are watching agentic payments closely as the technology expands the scope of autonomous financial transactions.