◆ Graph Engineering
Graph Engineering: evolve prompting into a system you can verify.Graph Engineering: evolve prompting into a system you can verify.

Free · Hands-on · Core labs need no API key

Graph Engineering

Evolve prompting into a system you can verify.

Build agent memory that survives the session: a temporal knowledge graph with a validation gate, cheap cached extraction, bounded retrieval that cites its evidence, and a routing policy that spends intelligence only where intelligence is needed.

Driven with Hermes Agent and GitHub Copilot CLI. Every command on this site was run before it was published.

Start the courseBrowse lessons

Read this first. This course exists because a claim went viral: "90% of Anthropic engineers shifted to agentic graphs. No more prompting." The percentages are unsourced and drifted between retellings, and the attached video was the Claude Code release demo, not a graph lecture. The architecture underneath is real and worth learning. This course keeps the architecture and audits the rest. The full audit is Lesson 0.

What you'll build

A temporal graph

Edges carry valid_from and valid_until, so it can answer where someone worked in 2024, not just today.

A validation gate

A closed relation vocabulary and endpoint consistency checks that reject incoherent edges before they enter memory.

Cheap extraction

A stable cached prefix and batched backfills, plus the honest measurement of when caching does and does not apply.

Bounded retrieval

A bounded neighbourhood around your seeds, temporally filtered, citing the edges it used.

An MCP server

Your graph as persistent agent memory in Hermes and Copilot CLI. Verified to load.

A routing policy

A config file your system reads, not a slogan you repeat.

The whole idea, in one table

WorkVolumeJudgmentConfiguration
ExtractionVery highLow, mechanicalCheap model, cached prefix, batched
Traversal and reasoningLowHigh, multi-hopFrontier model, small precise context

Everything in this course is a consequence of that split.

Course outline

  1. 0. What this is, and what it isn't
  2. 1. Setup: Hermes and Copilot CLIHands-on
  3. 2. Graphs versus vectors, and why the answer is both
  4. 3. The temporal model: time is not metadataHands-on
  5. 4. Extraction, and what it actually costsHands-on
  6. 5. Agentic workflow graphs: prompts with contracts
  7. 6. The validation gate: stopping bad data before it compoundsHands-on
  8. 7. The routing policy: make the tradeoff executableHands-on
  9. 8. Wiring the graph into Hermes and Copilot CLIHands-on
  10. 9. Measure it, do not assume itHands-on
  11. 10. The production track: Neo4j and GraphitiHands-on
  12. 11. What you actually learned