Your organization
never forgets.
Project memory. Org-wide pattern detection. Role expertise that compounds over time. Skills, rules, and best practices generated automatically — for every agent in your organization.
Memory at every level, always in sync.
Working, episodic, semantic, and procedural memory — all four CoALA types, across projects, roles, and your whole organization. Recalled when needed. Retained automatically.
Three levels of learning. One platform.
Project memory captures what happened. Org learning finds what repeats. Role learning distills how your best people think — and makes that expertise available to every agent.
Every project builds its own memory
Agents capture decisions, errors, and discoveries as they work. The next agent that touches the same project starts with full context — no re-introductions, no repeated mistakes.
cortex.recall("how do we handle migrations?")→ "archive before insert — incident #42"→ "rollback script: scripts/rollback.sh"
Patterns found across every project, automatically
Every night, Cortex reads all project memories and finds what repeats: recurring failures, proven practices, and procedures worth automating. No one has to write this report.
cortex.org_insights("softtek")→ pattern: "env vars missing in 6 projects"→ candidate: "db-migration-skill"
Expertise that compounds across people
Every night, Cortex observes how people in the same role work: what decisions an architect makes, on what criteria, and why. That knowledge accumulates into role-specific memory — so every agent can reason with the depth of your best people.
cortex.recall(role="architect")→ "prefer event-driven over polling"→ "document every ADR before merging"
Patterns become assets.
When Cortex finds something worth keeping, it generates three types of artifacts — each serving a distinct purpose, all reviewed before distribution.
// Agent invokes a generated skill/db-migration-skill→ archive step added automatically→ rollback script ready
Procedures agents invoke by name
When Cortex detects a recurring procedure, it generates a skill candidate for expert review. Once validated and approved, the skill is distributed as a plugin — available to every connected agent by name.
// Rule loads at session startrule: no-hardcoded-secrets→ active in Claude Code→ active in GitHub Copilot
Instructions that load automatically
When a recurring mistake is detected, Cortex generates a rule candidate. After expert review and approval, it is distributed as a plugin and loads at the start of every session — no invocation needed, compatible with any agent that supports instruction files.
// Promoted from 4 projectspractice: api-versioning-standard→ distributed to 12 projects→ status: approved
Standards proven across your organization
Practices that worked consistently across multiple projects are surfaced as candidates, reviewed by your team, and promoted as organizational standards. Every new project starts with the knowledge your best teams already have.
Works with the agents you already use.
Native plugin for Claude Code. MCP protocol for everything else. Rules generated by Cortex are plain instruction files — every major agent reads them.
Claude Code
Native plugin
# .claude/settings.json
"mcpServers": {
"cortex": {
"command": "cortex-mcp"
}
}Install the FRIDA CORTEX plugin. Memory recall and retention happen automatically at session start and end. Skills and rules are picked up from your project directory.
MCP-compatible agents
Model Context Protocol
# Any MCP client cortex_mcp_url: https://cortex.your-org.com token: cc_tok_abc123 scope: your-project
Any agent that supports MCP can connect to FRIDA CORTEX. Kiro, custom agents, and automated workflows all use the same protocol. One integration, all agents.
Generated rules work in every agent
Rules are plain instruction files — no vendor lock-in. When Cortex generates a rule from an org-wide pattern, it reaches every agent your team uses.
Connect once. Learn continuously.
From the first agent session to org-wide skills distributed automatically — every step runs without manual effort.
Connect your agents
Add the FRIDA CORTEX plugin to Claude Code, or connect any MCP-compatible agent. Setup takes minutes. Agents start contributing to project memory from their first session.
Agents learn as they work
Every session captures decisions, errors, and discoveries automatically. No manual documentation. No extra code. Project memory builds with every task completed.
Cortex synthesizes every night
Cross-project patterns are detected automatically. Recurring failures, proven practices, and automation candidates are identified across your entire organization.
Artifacts reach every project
Generated skills, rules, and best practices are reviewed once and distributed automatically. What one team discovers, every agent in your organization benefits from.
Human review before distribution
Generated skills, rules, and best practices go through a review step before reaching any project. Your team approves what gets distributed — Cortex handles everything else.
Four types of memory. Every one covered.
FRIDA CORTEX implements the full CoALA taxonomy, the most widely cited memory framework for LLM agents. Most production agents have one type. We give you all four.
Active task context, tool results, current PR diff — the LLM context window.
Ephemeral: cleared at end of session
"PR #247 had a hardcoded secret." "CORS is a false-positive on this framework."
Retained automatically via hooks
"User prefers FastAPI." "Service A depends on Service B." "Team Platform owns auth module."
Persistent: facts, entities, relations
Security checklist, deploy runbook, support ticket workflow — executable and reviewed by your team.
Generated from org patterns, reviewed before distribution
Most production agents
- ✓Working memory (context window)
- ~Some semantic memory via RAG
- ✗Episodic memory
- ✗Procedural memory
FRIDA CORTEX
- ✓Working memory — 60% compaction strategy
- ✓Episodic — automatic, hook-based, zero friction
- ✓Semantic — 4-strategy hybrid retrieval
- ✓Procedural — 4 mechanisms, self-improving
Give your agents a memory they can build on.
Project learning active from the first session. Org-wide skills and rules generated automatically from there.