Gigabrain blog: field notes on agent memory

Separate facts, past experiences and standing instructions so your agent can remember useful knowledge without turning every conversation into a rule.

Includes: Memory-record template

Memory systems3 min read

How to hand a project to a fresh AI agent

Build a small handoff packet with current state, evidence, open questions and action boundaries so a fresh agent can resume work without guessing.

Includes: Agent handoff template

Memory systems4 min read

Conflicting facts in AI memory: corrections, dates and source authority

Distinguish a correction from a contradiction, preserve effective dates and define who can resolve uncertainty before an agent picks a convenient answer.

Includes: Knowledge-source contract

Memory systems4 min read

Shared memory for AI agents: project scope, ownership and conflicts

Coordinate agents around source versions, edit ownership and reviewed merges so a shared knowledge base does not become a collection of silent overwrites.

Includes: Agent handoff template

Retrieval & graphs

Find the evidence, preserve its meaning and connect the facts that belong together.

Back to library ↑

Knowledge architecture7 min read

Build a knowledge graph from documents: a source-linked walkthrough

Build knowledge architecture that helps your AI agent find the right source, connect related facts and keep track of corrections.

Includes: Source-contract worksheet

Knowledge architecture4 min read

GraphRAG vs vector RAG: examples, costs and tradeoffs

Compare retrieval and knowledge graphs through concrete questions, evidence requirements and maintenance tradeoffs.

Includes: Source-contract worksheet

Knowledge architecture3 min read

What is RAG? Retrieval-augmented generation with a worked example

Follow a source from retrieval to an AI answer. Inspect a worked example, citation limits and a practical review worksheet.

Includes: Source-to-answer worksheet

Knowledge architecture3 min read

What is GraphRAG? How relationships add context to retrieval

Trace a question across connected documents. Learn how graph retrieval works, what it can miss and how to review its evidence.

Includes: Relationship-path worksheet

Retrieval & graphs3 min read

Entity resolution for knowledge graphs: aliases, duplicates and review

Resolve aliases without merging different people, clients or projects. Use stable identity, scoped names and a review queue for uncertain matches.

Includes: Knowledge-source contract

Retrieval & graphs4 min read

Document chunking for RAG: preserve context and source citations

Choose document boundaries around evidence, preserve headings and qualifiers, and evaluate splitting choices against the questions your agent must answer.

Includes: Retrieval evaluation kit

Retrieval & graphs5 min read

Hybrid search for AI memory: keyword, vector and graph retrieval

Understand what keyword and vector retrieval contribute, why their scores differ, and how to test a combined ranking on your own questions.

Includes: Retrieval evaluation kit

Retrieval & graphs4 min read

Knowledge graph schema for AI agents: entities, relationships and evidence

Choose a small set of meaningful relationships, define their direction and evidence rules, and test how the graph handles changes and uncertain links.

Includes: Knowledge-source contract

Retrieval & graphs4 min read

MCP and persistent agent memory: tools, storage and permissions

Understand what an MCP tool interface can expose, what a memory system still needs to do, and how to review a proposed retrieval contract.

Includes: Memory tool-contract worksheet

Retrieval & graphs3 min read

Multi-hop RAG: follow evidence across connected documents

Trace two evidence branches through original fictional sources. Inspect the executed graph paths, source receipts and a broken-bridge failure.

Includes: Multi-hop review

Reliability

Check answers against their sources and find out why a memory workflow failed.

Back to library ↑

Field test4 min read

Test AI agent memory across sessions: a reproducible checklist

A reproducible Gigabrain memory test: retrieve a known fact in a later process, inspect its source citation and check missing evidence.

Includes: Memory-evaluation workbook

Field notes3 min read

Why AI agent memory fails, and how to debug it

Trace memory failures through imports, retrieval, stale sources and reasoning. Use a practical fault-finding table and repeatable tests.

Includes: Memory-evaluation workbook

Reliability4 min read

How to verify AI citations against the original source

Audit whether an AI citation supports the exact claim, preserves qualifications and points to the current source rather than a plausible-looking reference.

Includes: Knowledge-source contract

Reliability4 min read

Stale RAG indexes: why an AI agent retrieves an old answer

Diagnose stale retrieval by separating original files, derived indexes and cached answers, then verify the repair without deleting useful source history.

Includes: Obsidian pilot checklist

Reliability4 min read

Evaluate AI memory retrieval: precision, recall and missing evidence

Build a query set, label relevant evidence, calculate precision and recall, and separate retrieval quality from the correctness of the final answer.

Includes: Retrieval evaluation kit

Reliability4 min read

How to delete AI memory from sources, indexes and backups

Map source files, extracted text, indexes, conversations and backups before removing information from an AI knowledge workflow.

Includes: Memory removal checklist

Reliability3 min read

When a document tries to give your AI agent instructions

Keep retrieved knowledge separate from tool permissions, and use a harmless prompt-injection exercise to inspect how your memory workflow handles untrusted text.

Includes: Memory-record template

Reliability3 min read

Back up your AI knowledge system and prove you can restore it

Identify the files that matter, separate source knowledge from rebuildable indexes, and run an isolated restore drill before relying on a backup.

Includes: Knowledge restore checklist

Workflows

Turn project knowledge into decisions, research and useful daily work.

Back to library ↑

Workflows4 min read

An AI second brain needs more than storage

Put your notes to work with an AI agent. Build a second-brain workflow that keeps sources, corrections and permissions clear.

Includes: Weekly-review template

Obsidian5 min read

Use an Obsidian vault as agent knowledge

Use your Obsidian notes in an AI memory workflow. Start with a small project, check source citations and understand where your data goes.

Includes: Obsidian pilot checklist

Workflows3 min read

Turn meeting notes into decisions your AI can use

Extract approved decisions, owners and open questions from meeting notes without turning suggestions or summaries into commitments.

Includes: Meeting decision template

Workflows3 min read

Keep client projects separate in AI memory

Design project scope, access boundaries and negative tests so similar names and shared folders do not mix one client's knowledge into another's work.

Includes: Client boundary checklist

Workflows3 min read

Build a research library your AI can cite responsibly

Capture sources, distinguish quotations from interpretations, and keep claim-level evidence so your research library supports useful answers instead of plausible summaries.

Includes: Research-source template

Workflows3 min read

Make an AI daily brief that tells you what changed

Build a source-backed daily brief from approved project notes, distinguish deadlines from suggestions, and flag stale or missing inputs before prioritizing work.

Includes: Daily brief template