Home

Welcome to the CortexPrism Wiki

CortexPrism (v0.53.0) is a self-hosted, open-source AI agent operating system that turns any LLM into a capable autonomous agent. It provides persistent memory, a rich tool ecosystem, sandboxed code execution, a full-featured web UI, and enterprise-grade security — all running locally on your machine or server.

Experience Levels

Beginner: core essentials and first-run paths.

Intermediate: operational workflows and deeper navigation.

Advanced: expert-level workflows and tuning.

Quick Links

• Getting Started — Installation and first-run guide
• Architecture — System design and component overview
• CLI Reference — All commands and their usage
• Configuration — Config file reference and environment variables
• Security Model — Parallax security, LLM supervisor, policy validator, vault, and audit
• LLM Providers — All 30 supported providers and configuration
• Plugin System — Extending CortexPrism with plugins
• REST API — HTTP and WebSocket API reference
• Contributing — How to contribute to the project
• Discord — Join the community on Discord
• Changelog — Release history and version notes

Key Features

• 30 LLM providers — Anthropic, OpenAI, Gemini, Groq, Ollama, DeepSeek, Cerebras, Perplexity, DeepInfra, Hyperbolic, MiniMax, Zhipu/GLM, Replicate, Cloudflare AI, and more
• OS Kernel — System call dispatcher with capability enforcement, RBAC (4 roles), process registry with parent-child tracking, resource accounting (token/cost per agent)
• Virtual filesystem — /cortex/agents/, /cortex/memory/, /cortex/config/, /cortex/workspace/ namespace abstraction
• OS API — Health dashboard, process tree viewer, capability/role inspector endpoints
• Built-in agents — 10 pre-configured agents (Assistant, Developer, Researcher, Architect, Analyst, Writer, DevOps, Security, Code Reviewer, QA/Tester) with specialized soul prompts
• Agent exec — One-shot non-interactive agent mode for CI/scripting (--json, --max-turns, --output)
• Config CLI — config get/set/unset/list/validate with dot-notation keys and encrypted credential storage
• 12 TUI slash commands — /model, /theme, /export, /review, /plan, /help, and more in interactive chat
• 60 built-in tools — File system, shell, web search, browser automation, code execution, GitHub, memory, code intelligence, computer use, skills, voice, dashboard, and more
• 5-tier memory — Episodic, semantic, reflection, graph, and skill memory with FTS5 + vector search
• Web UI — 58/58 backend systems covered with 30+ pages, dashboard, and real-time streaming
• A2A Protocol — Google Agent-to-Agent v1.0 compliance with JSON-RPC server and client
• MCP Gateway — Multi-server Model Context Protocol gateway with health checks, rate limiting, and approval workflow
• Distributed nodes — Hub ↔ Node architecture with tiered capabilities (root/sudo/unprivileged)
• Memori — Persistent agent checkpointing with full-state capture and resumable sessions
• Parallax + LLM supervisor — Multi-layer policy validator + LLM-based access control for sensitive data
• LLM security supervisor — 3-layer access control (classification → LLM review → human approval)
• Plugin system — ESM, MCP, and WASM plugins with sandboxed permissions
• Sub-agents — 11 typed child agents (Explorer, Coder, Researcher, Planner, Generalist, Security Auditor, Debugger, Architect, DevOps Engineer, Data Analyst, UI/UX Designer)
• Multi-user collaboration — Users, teams, API tokens, PBKDF2 password hashing, resource sharing, team join policies, and instance-to-instance federation
• Model Quartermaster — 6-signal prediction engine with adaptive EMA learning
• 9 channel adapters — Discord, Slack, Teams, Telegram, WhatsApp, Google Chat, Lark, Mattermost, RocketChat
• Voice pipeline — STT, TTS, VAD, and real-time audio streaming
• Code intelligence — Tree-sitter WASM parser for 12 languages with active tree-sitter parsers (40+ file extension mappings)
• Computer use — GUI automation via Xvfb with mouse, keyboard, and screenshot actions
• Browser automation — Headless Playwright-powered web interaction and screenshots
• Skills system — Self-learning skills with 6-state lifecycle, embedding retrieval, and trust tiering
• Eval system — Agent evaluation runner with regression detection and baselines
• Prompt Lab — Prompt template design, version tracking, and test runs
• PKM Assistant — Connects to Obsidian, Logseq, Notion, and Roam Research
• CACP — Cross-Agent Context Protocol for multi-agent session collaboration
• Triggers — Webhook, filesystem watcher, and git hook event triggers
• Workflow engine — DAG-based workflow execution with branching and parallelism
• TUI — Terminal UI for headless operation
• Update system — Binary and source update checking, installation, and rollback
• Observability — Prometheus metrics, OTLP tracing, Langfuse LLM observability
• Glossary — In-memory term registry for project-specific vocabulary
• Sandbox suite — Environment replication, workspace snapshots, dev env as code, bug reproduction studio
• Chrome Bridge — Real-browser automation with 60 tools via CDP (Chrome DevTools Protocol)
• AgentLint — Automated auditing of agent configs, tools, plugins, and prompts (29 checks)
• Responsible AI — Bias, fairness, and safety auditing for agent outputs (10 audit categories)
• Supply chain verification — Plugin integrity with SHA-256 hashing, signature verification, and malware scanning
• DLP Guard — 21-scanner data loss prevention for API keys, credentials, PII, PHI, PCI
• AI Guardrails — Pluggable content safety middleware with 5 built-in classifiers
• Codebase Pilot — Token-optimized context builder with AST-aware pruning
• Alcove — Private documentation search and browse over .cortex/data/docs/
• Remote sandbox backends — E2B and Daytona support beyond Docker and subprocess
• Distributed agent swarm — Cross-instance coordination via A2A protocol with fleet topology, resource aggregation, and directive dispatch
• Runtime tool forging — Agents create, test, and export custom tools at runtime with safety scanning
• Multi-agent orchestration — 6 strategies (sequential, parallel, debate, review-loop, hierarchical, graph) via orchestrate tool
• HEXACO personality system — 6-factor personality engine driving system prompts, memory bias, and routing hints
• Memory benchmarking — LongMemEval-S compatible benchmark suite with CI integration
• Checkpoint time-travel UI — Full timeline with resume and branch-from-here for memori checkpoints

License

Apache 2.0 — free for personal and commercial use.