# NEXO Brain > Open-source cognitive runtime with a shared brain for Claude Code, Codex, Claude Desktop, and other MCP clients (v5.0.0). NEXO Brain gives AI agents persistent memory using the Atkinson-Shiffrin cognitive architecture. Version 5.0.0 closes more of the loop between goals, decisions, outcomes, reusable skills, and public proof, while keeping the local shared-brain model across supported clients, configurable terminal and automation backends, recovery-aware background jobs, startup preflight, runtime self-healing, multimodal memory refs, pre-compaction auto-flush, a public claim wiki, readable memory export, richer user-state adaptation, long-horizon overnight analysis, and 150+ MCP tools. ## Core Product Model - Shared brain is the baseline. - Interactive client and automation backend are configurable separately. - Claude Code remains the recommended path because it has the deepest hook and headless automation integration. - Codex is supported as both terminal client and automation backend. - Claude Desktop can share the same brain through MCP. ## Current Release Focus - Goal profiles and Decision Cortex v2 shape higher-stakes recommendations instead of leaving strategy selection to raw prompting. - Outcomes, impact scoring, and structured pattern capture connect action, result, and future prioritization more tightly. - Repeated winning patterns can seed or promote reusable skills, while poor evidence can demote or retire them. - Public proof now combines the LoCoMo memory result with a broader operator runtime pack so the website reflects more than one memory benchmark. ## Key Features - Three-store memory: Sensory Register, Short-Term Memory, Long-Term Memory - Semantic search with local embeddings and optional HNSW acceleration - Claim graph, trust scoring, metacognitive guard, cognitive dissonance detection - Multimodal memory references for images, audio, video, PDFs, and other non-text artifacts - Structured auto-flush before compaction so actionable continuity survives compression - Public claim/wiki tools with evidence, freshness, verification state, and linting - Readable markdown memory export for auditability - Stronger inspectable user-state model beyond shallow sentiment heuristics - Public retrieval knobs for hybrid weighting, decomposition, dream exclusion, and dormant-memory handling - Explicit backend contract/status for newer memory layers - Episodic memory, reminders, learnings, followups, and session continuity - 24-hour hot context for recent operational continuity across sessions and channels - Transcript fallback tools for recent Claude Code / Codex conversations when the memory layer is too thin - Live system catalog for tools, plugins, skills, scripts, crons, projects, and artifacts - Context continuity across Claude Code compaction via hooks - Shared-brain client sync for Claude Code, Codex, and Claude Desktop - Managed bootstrap parity for `~/.claude/CLAUDE.md` and `~/.codex/AGENTS.md` - Managed Codex global bootstrap/model sync via `~/.codex/config.toml` - Managed Codex `mcp_servers.nexo` sync inside `~/.codex/config.toml` - Existing-install backfill so older Codex-linked runtimes actually receive that managed global bootstrap/model sync during `nexo update` - Installed-runtime bootstrap hotfix so existing users receive those managed bootstrap updates correctly - Bootstrap operator-name fallback so managed `CLAUDE.md` / `AGENTS.md` never render with blank identity headings - Explicit `CORE` / `USER` contract so updates can refresh product rules without overwriting operator instructions - Configurable `nexo chat` terminal client - Configurable automation backend with client-specific model profiles - Automatic HyDE / spreading activation defaults that stay conservative for exact lookups - Retrieval explanations with confidence + auto-strategy hints - Per-memory `stability` / `difficulty` personalization on top of Ebbinghaus decay - Deep Sleep transcript parity across Claude Code and Codex session stores - 60-day long-horizon Deep Sleep blend for recurring themes and older/current cross-domain connections - Weighted project-priority signals plus weekly/monthly Deep Sleep summary artifacts - Periodic Deep Sleep summaries with protocol compliance, engineering-loop output, project pulse, and trend-vs-previous data - Semantic Deep Sleep followup deduplication plus learning consolidation outcomes: duplicate, reinforcement, contradiction review, or genuinely new - Deep Sleep engineering followup backfill so recurring patterns turn into concrete scripts, hooks, checklists, and guardrails instead of only passive observations - Runtime doctor parity audits for recent Codex sessions and Claude Desktop shared-brain metadata - Runtime doctor audits weekly protocol compliance plus release artifact sync drift - Dashboard narratives for what matters now, what is drifting, and what is improving - Public `scripts/verify_release_readiness.py` validator for release discipline - Pending Codex launcher fixes included: client selection, launch mode, last terminal choice, interactive flags - Runtime CLI with `nexo chat`, `nexo update`, `nexo doctor`, and personal script management - Self-healing runtime with managed schedule reconciliation and KeepAlive daemon support - Personal scripts registry, dashboard, overnight learning, and recovery-aware jobs - Claude Code plugin packaging, OpenClaw package, and ClawHub skill ## Recommended Defaults - Claude Code: recommended primary client - Claude Code model profile: Opus 4.6 with 1M context - Codex model profile: GPT-5.4 xhigh ## Installation ```bash npx nexo-brain ``` ## Update Existing Installations ```bash nexo update ``` ## Re-sync Clients Later ```bash nexo clients sync ``` ## Main Commands - `nexo chat` — open the configured terminal client - `nexo update` — sync the installed runtime - `nexo doctor` — audit and repair runtime health - `nexo scripts list` — inspect personal scripts ## Documentation - Website: https://nexo-brain.com - Changelog: https://nexo-brain.com/changelog/ - Blog: https://nexo-brain.com/blog/ - Docs: https://nexo-brain.com/docs/ - Install flow: https://nexo-brain.com/docs/install-flow/ - GitHub: https://github.com/wazionapps/nexo - npm: https://www.npmjs.com/package/nexo-brain ## Compatibility NEXO Brain exposes its capabilities as MCP tools. It is compatible with MCP-capable environments and currently has explicit shared-brain flows for Claude Code, Codex, and Claude Desktop. OpenClaw is also supported through the published package. ## Contributing Issues and PRs are welcome, especially for client parity, installer UX, public docs, compatibility validation, and real-world runtime reliability. ## License AGPL-3.0. Built by WAzion.