Labsco
abbacusgroup logo

Cortex

★ 20

from abbacusgroup

Ontology driven knowledge system with formal OWL-RL reasoning, SPARQL graph + SQLite dual store, and self-improving memory tiers. 22 MCP tools for capture, search, reasoning, graph operations, and diagnostics. Local-first.

🔥🔥🔥✓ VerifiedAccount requiredAdvanced setup
#analytics#database#devops#git#maps#memory

Cortex

Cognitive knowledge system with formal ontology, reasoning, and intelligence serving.

Cortex captures knowledge objects (decisions, lessons, fixes, sessions, research, ideas), classifies them with an OWL-RL ontology, discovers relationships, reasons over the graph, and serves intelligence through hybrid retrieval.

CLI Commands

Command Description cortex setup Interactive setup wizard cortex init Initialize data directory and stores (deprecated — use cortex setup) cortex install Install background services (macOS/Linux) cortex uninstall Remove background services cortex register Register MCP server with Claude Code cortex capture Capture a knowledge object cortex search Hybrid keyword + semantic search cortex read Read object in full cortex list List objects with filters cortex status Health and counts cortex context Briefing mode (summaries) cortex dossier Entity-centric intelligence brief cortex graph Show object relationships cortex synthesize Cross-document synthesis cortex entities List resolved entities cortex serve Start MCP or HTTP server cortex dashboard Start web dashboard cortex backup Backup data directory to archive cortex restore Restore from backup archive cortex doctor Diagnostics: check, unlock, logs, repair cortex pipeline Re-run intelligence pipeline on an object cortex reason Run advanced reasoning (contradictions, patterns, gaps) cortex import-v1 Import from Cortex v1 database cortex import-vault Import from Obsidian vault

MCP Tools

26 tools for AI agent integration. Localhost-bound HTTP exposes all; non-localhost binds expose only the public set.

Public: cortex_search, cortex_context, cortex_dossier, cortex_read, cortex_capture, cortex_link, cortex_feedback, cortex_graph, cortex_list, cortex_classify, cortex_pipeline

Admin (localhost only): cortex_status, cortex_synthesize, cortex_delete, cortex_delete_entity, cortex_update, cortex_unlink, cortex_reason, cortex_query_trail, cortex_graph_data, cortex_list_entities, cortex_export, cortex_import, cortex_safety_check, cortex_debug_sessions, cortex_debug_memory

Architecture

Cortex runs as a single MCP HTTP server that owns the graph store. Claude Code, the dashboard, the CLI, and the REST API are all HTTP clients of that one server.

Copy & paste — that's it
┌───────────────┐ ┌────────────┐ ┌─────────────┐
│ Claude Code │ │ Dashboard │ │ CLI │
│ (MCP client) │ │ (browser) │ │ (terminal) │
└───────┬───────┘ └─────┬──────┘ └──────┬──────┘
 │ │ │
 │ HTTP JSON-RPC │ HTTP MCP │ HTTP MCP (default)
 │ │ │ direct (--direct)
 ▼ ▼ ▼
 ┌──────────────────────────────────────┐
 │ cortex serve --transport mcp-http │
 │ (canonical MCP HTTP server) │
 │ PID-locked owner of graph.db │
 └──────────────────────────────────────┘
 │
 ▼
 ┌─────────────────────────────┐
 │ ~/.cortex/ │
 │ graph.db (Oxigraph) │
 │ cortex.db (SQLite WAL) │
 └─────────────────────────────┘

  • Ontology: OWL-RL formal ontology with 8 knowledge types and 8 relationship types

  • Storage: Oxigraph (RDF/SPARQL) + SQLite (FTS5/BM25) dual-write

  • Pipeline: Classify → Extract entities → Link → Enrich → Reason

  • Retrieval: Hybrid keyword + semantic + graph-boosted ranking

  • Serving: 5 presentation modes (briefing, dossier, document, synthesis, alert)

  • Transports: MCP (stdio + HTTP), REST API, Web Dashboard

Service Management

Copy & paste — that's it
# Install both MCP server and dashboard as background services
cortex install

# Install only the MCP server
cortex install --service mcp

# Remove all services
cortex uninstall

On macOS, this creates LaunchAgent plists (auto-start on login, auto-restart on crash). On Linux, this creates systemd user units.

Raw templates are available in deploy/ for manual setup.

--direct escape hatch

By default, CLI commands route through the running MCP server. If the server is down:

Copy & paste — that's it
cortex --direct list # bypass MCP, open store directly
cortex --direct pipeline --batch # required for bulk SQL operations

Bootstrap commands (init, setup, import-v1, import-vault) always run directly.

Docker

Copy & paste — that's it
docker compose up -d
# Server at http://localhost:1314

Knowledge Types

decision, lesson, fix, session, research, source, synthesis, idea

Relationship Types

causedBy, contradicts (symmetric), supports, supersedes (transitive), dependsOn, ledTo (inverse of causedBy), implements, mentions

Privacy

Cortex stores all data locally. No telemetry, no analytics, no phone-home. If you configure an LLM provider (via CORTEX_LLM_API_KEY), object content may be sent to that provider for classification and reasoning. Embeddings are computed locally by default using sentence-transformers.

License

Copyright (c) 2026 Abbacus Group. Licensed under the MIT License.

Trademark Notice

Cortex is a project of Abbacus Group and is not affiliated with any other product named Cortex.