Ingest documents. Extract knowledge graphs. Store temporal entities and relations. Query with bounded traversal. Your agents remember what they learned across sessions.
A durable pipeline that ingests, parses, chunks, extracts, and reconciles. Automatic retries on failure.
Upload PDF, Markdown, HTML, or plain text. Size, type, and duplicate validation.
Extract text content. Split into semantic chunks preserving context boundaries.
LLM-powered entity, relation, and evidence extraction from each chunk.
Deduplicate entities. Merge temporal versions. Build the knowledge graph.
Works with Claude Code, OpenCode, and Hermes through MCP protocol. One bridge, multiple harnesses.
Anthropic's coding agent. Configure via .claude/mcp.json.
SVC Labs' coding agent. Configure via opencode.json.
Agent framework. Configure via hermes.config.json.
The skill tells agents what to do. The bridge connects them. The core stores everything.
Tells AI agents how and when to use Agent Memory. Defines the recall, reorient, and persist protocol. Not a runtime — a behavioral contract.
View repoStateless MCP bridge connecting Claude Code, OpenCode, and Hermes to OpenGraphMemory REST APIs. 20 tools, 3 permission profiles. Published on PyPI.
View repoKnowledge graph extraction, temporal storage, and Agent Memory API. FastAPI + PostgreSQL. The backend that powers everything.
View repoStructured memory with temporal supersession. Agents record what they learned, not what they said.
Immutable observations of what happened. Errors encountered, fixes applied, patterns discovered.
Hypotheses tested within an episode. What was tried, what happened, what was the result.
Verified results with verifiers. A test passes, a build succeeds, a deploy completes.
Distilled knowledge from verified outcomes. Durable conclusions that supersede earlier understanding.
Query the knowledge graph at any point in time. See how entities and relations evolved.
Read-only by default. Personal-safe for writes. Memory-curator for curation governance.