Tagged: Knowledge-Graph

Knowledge graph evaluation usually measures retrieval. Scoring 12 real decisions against the answer you would have reached anyway shows more.

Compare Palantir Foundry with a focused open-source context layer: scope, costs, governance and a runnable delivery-blocker demo.

A temporal knowledge graph stores every fact with when it held and when it was recorded. How it works for AI agents, how Graphiti builds it, and when to use one.

Graphiti vs Mem0 in September 2026: how each stores and updates facts, what runs self-hosted, what the benchmark fight says, and which fits your agents.

Graphiti Local is an open-source, local-first temporal knowledge graph on Graphiti: six read-only MCP tools, a kg CLI, and writes only after a person approves.

Running Graphiti in production: FalkorDB or Neo4j, Azure OpenAI or Ollama, embedding widths, version pins, ingestion cost and the failures to watch for.

The failure no dead-letter queue catches: an agent is certain which customer a record belongs to, and it is wrong. Renames, subsidiaries, product names as account names, per-field authority, corrections as aliases, and why silence is not decay.

Case study: a context layer for AI agents on a temporal knowledge graph. Sourced answers from chat, tickets, wiki and CRM, external actions authorized before execution.

A human gate on agent memory writes is now consensus. The volume reaching it is the unsolved half. Three routing rules that keep the queue readable.

Knowledge graph agent memory comes in two kinds: one half is derivable from your code, one half only people can author. Mixing them wastes both.