Service
Glossary
Alphabetical index of every knowmind term. Where a concept or reference page exists, the entry links to it.
- API Token
- Personal Access Token of the form kmt_… that authenticates HTTP calls to knowmind. Passed in the Authorization header as Bearer. Available from the Business API plan.
- Read more →
- Audit log
- Record of every security-relevant action (token creation, BYOK changes, plan changes, memory mutations). Retention depends on the plan (30 days to 5 years).
- Read more →
- Bi-temporal
- Modelling with two time axes: valid when (valid_from, valid_to) and known when (created_at, updated_at). Enables point-in-time recall.
- Read more →
- BM25
- Full-text ranking function in PostgreSQL. One of the stages in the recall pipeline. Finds literal matches quickly and is robust to typos.
- Read more →
- BYOK KeyBYOK = Bring Your Own Key
- Your own AI-provider key that you store in knowmind. knowmind uses it for internal enrichment and indexing. The key is stored AES-256-GCM encrypted.
- Read more →
- Chunk
- A content section of a document, typically 500 tokens with 50 tokens overlap. Own vector per chunk.
- Read more →
- Cross-encoder rerank
- Last stage of the recall pipeline. A model that reads question and candidate content together and returns a precise score.
- Read more →
- Dashboard
- The browser UI under /dashboard. Allows search, save, plan management, BYOK keys, API tokens and audit inspection.
- DCR — Dynamic Client RegistrationRFC 7591
- OAuth standard by which custom connectors register themselves automatically with the authorization server (knowmind).
- Read more →
- Embedding
- Numeric representation of a text section as a vector. knowmind uses multilingual-e5-large with 1024 dimensions. Foundation of semantic search.
- Read more →
- HMAC
- Hash-based Message Authentication Code. knowmind signs webhook bodies with HMAC-SHA256 and a secret generated per subscription.
- Read more →
- HNSWHierarchical Navigable Small World
- Approximate nearest-neighbour index in pgvector. Keeps vector search fast even with millions of embeddings.
- Integration
- Connection of an external tool (Claude Code, ChatGPT, Cursor) to knowmind via MCP or OAuth.
- Read more →
- JSON-RPC 2.0
- Call format for MCP. Requests carry jsonrpc, id, method, params. Responses carry result or error.
- Read more →
- KEK — Key Encryption Key
- Per-workspace derived key that encrypts BYOK provider keys. Derived from the master key via HKDF-SHA256.
- Read more →
- Knowledge Graph
- Neo4j-backed graph of memories and their typed relations. Enables graph hops in the recall pipeline.
- Read more →
- Magic Link
- Sign-in mail with a one-time link valid for 15 minutes. Replaces the classic password.
- MCP — Model Context Protocol
- Open standard for connecting tools to language models. knowmind is an MCP server at https://knowmind.de/api/mcp/v1.
- Read more →
- Memory
- The smallest knowledge unit in knowmind. Has a title, content, type (semantic, episodic, procedural, reference), optional tags and source. On creation it is chunked, embedded, and linked in the Knowledge Graph. German UI: "Erinnerung".
- Read more →
- Memory Pack
- A system prompt that tells an AI when and how to use knowmind. Stored in CLAUDE.md, ChatGPT custom instructions or workspace prompts.
- Read more →
- Memory Store
- Conceptual term for the content of a workspace. In billing terms the unit is the "Workspace", in conceptual terms it is the "Memory Store".
- Multilingual-e5-large
- Embedding model used by knowmind by default. 1024 dimensions, multilingual (100+ languages), MIT licence.
- OAuth 2.1
- Authentication standard for custom connectors. knowmind is an OAuth 2.1 provider with Dynamic Client Registration and PKCE.
- Read more →
- Personalized PageRank
- Graph algorithm that propagates relevance weights through the graph from anchor nodes. One of the stages in the recall pipeline.
- pgvector
- PostgreSQL extension for vector operations and ANN indexes. knowmind uses it for vector search.
- PKCE — Proof Key for Code Exchange
- OAuth extension that protects public clients (browser, mobile apps) against code interception.
- Plan
- Subscription tier of the knowmind platform: Private, Pro, Business, Business API, Enterprise. Internal code label: "plan".
- Read more →
- Recall
- The act of finding a relevant memory for a question. knowmind uses hybrid recall with BM25, vector, graph and cross-encoder.
- Read more →
- Recall-at-time
- Point-in-time recall. Returns only memories that were valid at the given moment.
- Read more →
- Reciprocal-Rank Fusion — RRF
- Algorithm for fusing multiple rankings. knowmind uses RRF with k=60 in the recall pipeline.
- RLS — Row-Level Security
- PostgreSQL mechanism that enforces tenant separation at the database level. The knowmind app role holds NOBYPASSRLS — we cannot defeat the separation either.
- Read more →
- Scope
- Permission level of a token: read (recall, stats), write (create, relations), admin (plan actions).
- Read more →
- SHACLShapes Constraint Language
- W3C standard for schema validation in RDF graphs. knowmind validates every new relation against SHACL shapes.
- Streamable HTTP
- MCP transport using Server-Sent Events. Required by native custom connectors in claude.ai and ChatGPT. Activated via Accept: text/event-stream.
- Read more →
- SUPERSEDES
- Relation type that links a new version of a memory to the older one it replaces. Set automatically by knowmind.update_fact.
- Read more →
- Tenant
- Technical synonym for Workspace. In documentation the preferred term is "Workspace".
- Webhook
- Outbound HTTP POST from knowmind to a URL you register, fired by an event. HMAC-signed, retry-safe. Available from Business API.
- Read more →
- Workspace
- The tenant-isolated unit in knowmind. Every account has one; Business plans can invite additional members. Contains memories, relations, documents, tokens, BYOK keys and audit log.
- Read more →