Cognition compare
Cognition vs Supermemory
Supermemory (public docs) ingests documents and answers with hybrid search over stored memories. Cognition stores approved procedural skills and returns matching skill bodies when the agent calls cognition_assist. Other authorized ingestion paths serve separate purposes.
How Cognition works in the product
Not a wiki lookup: behavior shipped in this codebase today.
- Agents call one MCP entry point, cognition_assist; the runtime picks context, replay, install, or bundle responses instead of making the host orchestrate separate memory tools.
- Approved skills live in clo_skills with author, visibility, freshness, and structured trace fields; assist returns markdown bodies directly in the tool result.
- Skill saves require user approval. Ownership, privacy, and policy checks govern what can be saved or shared.
- Matching blends lexical ranking, trigger overlap, embeddings, project scope, and share-acceptance before a skill is shown.
- Apply-time governance compares the skill's captured risk posture to the active project and can block or ask before the body is applied.
- Eval tools (retrieval eval, run_skill_eval, optimize_skill) measure or tune skills, but promotion still flows through cognition_save_skill with approval.
Cognition is for
You and your coding agents
Coding-agent users who want approved procedures delivered inside Cursor, Codex, Claude Code, and Gemini via MCP, with author attribution and outcome receipts.
Supermemory is for
Their primary user
Teams building apps that need document ingestion, namespace scoping and hybrid retrieval (per Supermemory public API documentation).
Side by side
| Aspect | Cognition | Supermemory |
|---|---|---|
| Core object | Author-owned SKILL.md in clo_skills with trigger metadata | Documents and memory entries with embeddings (public Supermemory types) |
| Primary API shape | MCP tools: assist, save_skill, report_skill_outcome, forget | REST document add and hybrid search endpoints (public docs) |
| Retrieval delivery | Full skill body pushed in assist tool result for the host agent | Ranked memory or chunk passages from search response |
| Persistence gate | Hard stop without approved_by_user; privacy and policy lint | Ingestion and update pipelines defined by the memory product |
| Multi-agent sync | Same brain address and cog_* key across hosts via installers | Supermemory provides API integrations, agent skills, and MCP support |
| Forgetting / decay | freshness_band, teachability_state, and outcome-driven surfacing rules | Documented forgetting operations exclude selected memories from search while preserving them |
When to pick Supermemory
- You need a general-purpose memory API with hybrid search as the main interface.
- Your product ingests PDFs, files, and connectors at scale (Supermemory documents this surface).
- You are not standardizing on Cognition's MCP coding-agent install path.
How they work together
Supermemory can remain the document and passage retrieval layer for a custom app. Cognition fits when you want a governed SKILL.md layer that coding agents pull through MCP with explicit approval and outcome reporting.
Try Cognition free for two weeks
Memory and judgment for you and your agents. Install in about two minutes for Cursor, Codex, Claude Code, or Gemini. Then $9.99/mo, cancel anytime.
Sources (public)