GNO vs GBrain

Positioning

GBrain gives an agent a brain. The corpus is markdown in a git repo, backed by Postgres/pgvector for hybrid search. A zero-LLM extractor wires people and companies into a knowledge graph, and a background job queue enriches the brain overnight. After conversations the agent writes pages on its own. Single-user, agent-facing, with 30+ MCP tools, voice capture, and published corpus and retrieval figures on its own benchmark. The self-wiring graph and the overnight loop are genuinely novel.

GNO is a knowledge engine over folders you already have and over notes your agents capture into them. One SQLite index covers markdown, code, PDFs, Office documents, and portable mail, calendar, transcript, JSONL, and browser exports. Agents write through gno capture with explicit --source-kind provenance and opt-in MCP write tools; gno watch, gno daemon, and a scheduled gno index keep the corpus current. That loop runs because the operator wired it: GNO does not write back after a conversation on its own and runs no overnight self-enrichment. On the way out it returns deterministic Context Capsules, verified answers that abstain below full claim support, a change journal, and per-collection fail-closed egress, with a human workspace on the same index.

Both products are agent-facing write-and-retrieve systems. The split is who owns the corpus contract and the overnight agenda: GBrain accumulates ambiently on its schedule; GNO accumulates deliberately on yours.

Key differences

• Agenda — GBrain writes after conversations and enriches overnight. GNO capture, watch, daemon, and scheduled index run because the operator, or a skill the operator installed, started them. Ambient accumulation against deliberate accumulation.
• Corpus — GBrain is a markdown brain in git with its own page shape. GNO indexes any folder: markdown, code, PDFs, DOCX, XLSX, PPTX, and portable exports as read-only records. Existing files and agent captures land in the same index.
• Graph — GBrain self-wires people and companies with zero-LLM entity extraction. GNO’s graph connects wiki links, markdown links, similarity edges, and typed doc_edges, traversable from CLI, Web UI, REST, and MCP.
• Provenance — gno capture writes structured source frontmatter with a receipt; the MCP capture tool uses the same contract under the shared write lock. GBrain’s post-conversation write is the memory itself; GNO’s capture is an explicit filing.
• Read-back — GNO compiles Context Capsules with exact line spans, content hashes, one token budget, and declared gaps, and verified Ask withholds a draft unless every substantive claim is supported. GBrain publishes retrieval figures on its own corpus; this page does not put a competing score next to them.
• Egress and workspace — Each GNO collection carries a fail-closed local, LAN, or remote policy, and derived evidence inherits the strictest source. A human workspace with browse, editing, native PDF pages, and the graph sits on the same index. GBrain is agent-facing.

Running both

Point GNO at GBrain’s markdown directory. GBrain keeps its write-back, entity graph, and overnight enrichment. GNO indexes that brain beside your PDFs, Office documents, code, and exports, and serves the combined corpus to humans and other agents with Capsules, verified answers, and egress policy. They do not fight over the files.

gno setup ~/path/to/gbrain --name gbrain
gno query "who did we meet at the last offsite"
gno ask "what did we decide about pricing" --verify --show-sources

MCP writes stay opt-in (gno mcp --enable-write). Capture does not auto-embed; run gno index or gno embed when vector search must see new pages. The daemon watches for file changes; it does not replace GBrain’s maintenance loop.