{ "schema": "lxkeys.world.entry@2.2.1", "language": "fr", "entry": { "id": "ENT-00003000", "world_id": "ENT-00003000", "status": "published", "name": "Muse", "aliases": [ "Meta Muse" ], "kind": "Agents", "type": "AI Agent", "subtype": "Personal AI Agent", "classification": "Autonomous / Agentic AI", "creator": "Meta", "organization": "Meta", "origin_context": "Meta personal AI agent ecosystem and Meta Superintelligence Labs model stack", "first_public_appearance": "2026-09-08", "current_status": "Active", "official_website": "https://muse.ai/", "image": "", "image_meta": [], "short_description": "Muse est l’agent d’IA personnel de Meta, conçu pour agir de manière proactive sur des objectifs et des tâches, utiliser des applications connectées et le web, et poursuivre des travaux en arrière-plan dans une Muse Secure VM dédiée.", "public_description": "Muse est un agent d’IA personnel présenté par Meta le 8 septembre 2026. Il est conçu pour dépasser l’assistance conversationnelle tour par tour en planifiant et en exécutant des tâches pour le compte de l’utilisateur, en poursuivant des travaux en arrière-plan, en utilisant des services connectés et en revenant vers l’utilisateur lorsqu’une action conséquente exige son approbation.", "technical_description": "Muse fonctionne dans une Muse Secure VM dédiée avec un environnement d’agent isolé, un navigateur, un système de fichiers et un terminal. Sentinel contrôle les permissions des connecteurs et les sorties réseau, tandis que les identifiants sont stockés hors de l’environnement principal de l’agent. Muse utilise la famille de modèles Muse Spark ; la documentation de sécurité publiée au lancement identifie spécifiquement Muse Spark 1.3 pour les workflows de navigation et de résistance aux injections de prompt.", "capabilities": [ "Personal goal planning and multi-step task execution", "Proactive suggestions and long-running background work", "Web browsing, navigation and form completion", "Computer-use and tool-use workflows", "Connected-app workflows through scoped connectors", "Email and calendar workflows where authorized", "Travel research, booking workflows and transaction preparation", "Document, PDF, web-page and interactive artifact creation", "Custom tool and connector creation using a dedicated runtime", "Multi-agent and subagent coordination", "Persistent conversational memory and user-controlled personalization", "Direct interaction through the Muse app, web interface and WhatsApp", "Human-in-the-loop approval for sensitive or consequential actions", "Auditable activity and permission history" ], "limitations": [ "At launch, general availability was rolling out in the United States on iOS, Android and the web; geographic availability may expand over time.", "AI-glasses support was announced as coming soon rather than generally available at launch.", "Actions depend on the services, connectors, permissions and account access explicitly granted by the user.", "Sensitive or consequential actions can require explicit user approval and may not proceed autonomously.", "Like other AI agents, Muse can make mistakes and can encounter adversarial or misleading external content; Meta documents defense-in-depth controls rather than claiming zero residual risk.", "Background work and personalized memory increase the importance of permission review, auditability and careful handling of connected personal data.", "Security, privacy and safety properties in this record are documented from Meta’s first-party launch materials and should not be interpreted as independent third-party assurance.", "Specific model versions, connectors, payment integrations and product availability can change as Meta updates the service." ], "timeline": [ { "date": "2026-04-08", "title": "Muse Spark model family introduced", "description": "Meta Superintelligence Labs introduced Muse Spark, the first model in the Muse family and an early technical foundation for Meta’s agentic and personal-superintelligence direction.", "source_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "verification_status": "source_backed_official" }, { "date": "2026-07-09", "title": "Muse Spark 1.1 expanded agentic capabilities", "description": "Meta introduced Muse Spark 1.1 with stronger tool use, computer use, multimodal reasoning and orchestration across external apps and services.", "source_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "verification_status": "source_backed_official" }, { "date": "2026-09-02", "title": "Muse Spark 1.3 released", "description": "Meta released Muse Spark 1.3 with improvements for longer-horizon agentic workflows, multitasking, coding, user collaboration and consequential-action awareness. Launch-era Muse security documentation identifies Muse Spark 1.3 in the agent’s browser and prompt-injection defenses.", "source_url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "verification_status": "source_backed_official" }, { "date": "2026-09-08", "title": "Muse publicly introduced", "description": "Meta introduced Muse as a personal AI agent designed to proactively work toward goals, execute tasks on a person’s behalf and continue work in the background.", "source_url": "https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/", "verification_status": "source_backed_official" }, { "date": "2026-09-08", "title": "Muse Secure VM and Sentinel architecture documented", "description": "Meta published launch documentation describing Muse Secure VM, isolated runtime boundaries, credential separation, Sentinel-controlled permissions and network egress, and human-in-the-loop approvals.", "source_url": "https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse", "verification_status": "source_backed_official" }, { "date": "2026-09-08", "title": "Initial United States rollout began", "description": "Muse began rolling out in the United States on iOS, Android and muse.ai. Meta also documented direct use through WhatsApp and announced future AI-glasses support.", "source_url": "https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/", "verification_status": "source_backed_official" }, { "date": "2026", "title": "Muse Confidential VM announced for later in 2026", "description": "Meta announced a planned Muse Confidential VM in which the VM, user data and Muse conversations would be encrypted with a key held by the user.", "source_url": "https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/", "verification_status": "source_backed_official_planned" } ], "sources": [ { "label": "Introducing Muse: The World’s First Personal AI Agent Built for Everyone", "url": "https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/", "source_type": "Official Product Release", "verification_status": "verified", "authority": "OFFICIAL", "role": "primary", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" }, { "label": "Muse official product site", "url": "https://muse.ai/", "source_type": "Official Product Source", "verification_status": "verified", "authority": "OFFICIAL", "role": "product", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" }, { "label": "How We Built Safety Into Muse", "url": "https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse", "source_type": "Official Technical Documentation", "verification_status": "verified", "authority": "OFFICIAL", "role": "technical_documentation", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" }, { "label": "How We Designed Muse", "url": "https://introducing.muse.ai/", "source_type": "Official Product Design Documentation", "verification_status": "verified", "authority": "OFFICIAL", "role": "documentation", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" }, { "label": "Introducing Muse Spark 1.3", "url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "source_type": "Official Model Release", "verification_status": "verified", "authority": "OFFICIAL", "role": "model_lineage", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" }, { "label": "Introducing Muse Spark 1.1", "url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "source_type": "Official Model Release", "verification_status": "verified", "authority": "OFFICIAL", "role": "model_lineage", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" }, { "label": "Introducing Muse Spark", "url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "source_type": "Official Model Release", "verification_status": "verified", "authority": "OFFICIAL", "role": "model_lineage", "update_enabled": true, "authority_basis": "official-source-review-2026-09-14" } ], "relationships": [ { "target": "Meta", "type": "Organization Context", "description": "Meta created, operates and publishes Muse as part of its consumer AI product portfolio.", "evidence_level": "documentary", "source_url": "https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/" }, { "target": "Meta AI", "target_id": "ENT-00000033", "type": "Related AI Product", "description": "Muse is a distinct personal-agent product in Meta’s AI portfolio, while Meta AI is Meta’s general-purpose assistant; both use the broader Muse model ecosystem.", "evidence_level": "documentary", "source_url": "https://ai.meta.com/" }, { "target": "Muse Spark 1.3", "type": "Model Infrastructure Relation", "description": "Muse is powered by the Muse Spark model family; Meta’s launch-era security documentation identifies Muse Spark 1.3 in browser-driving and prompt-injection-resistant agentic workflows.", "evidence_level": "documentary", "source_url": "https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse" }, { "target": "Muse Secure VM", "type": "Runtime Infrastructure", "description": "Muse runs in a dedicated Muse Secure VM that stores user data and connected-service credentials and isolates the agent runtime from security-sensitive services.", "evidence_level": "documentary", "source_url": "https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse" }, { "target": "WhatsApp", "type": "Distribution Surface", "description": "Meta documents Muse as directly usable through WhatsApp in addition to the Muse app and web interface.", "evidence_level": "documentary", "source_url": "https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/" } ], "tags": [ "AI Agent", "Personal AI Agent", "Agentic AI", "Meta", "Muse", "Muse Spark", "Muse Secure VM", "Computer Use", "Background Agent", "Multi-Agent Systems", "Personal Superintelligence", "WhatsApp", "Human-in-the-Loop", "Intelligent Entity" ], "facts": [], "canonical": [], "source_state": [], "i18n": { "en": { "short_description": "Muse is Meta’s personal AI agent, designed to work proactively on goals and tasks, operate across connected apps and the web, and continue background work on a dedicated Muse Secure VM.", "public_description": "Muse is a personal AI agent introduced by Meta on September 8, 2026. It is designed to move beyond turn-by-turn assistance by planning and executing tasks on a person’s behalf, continuing work in the background, using connected services and returning for approval when consequential actions require the user’s decision.", "technical_description": "Muse runs in a dedicated Muse Secure VM with an isolated agent runtime, browser, filesystem and terminal. Sentinel governs connector permissions and network egress, while credentials are stored outside the main agent runtime. Muse uses the Muse Spark model family, with launch-era security documentation specifically identifying Muse Spark 1.3 for browser-driving and prompt-injection-resistant agentic workflows." }, "fr": { "short_description": "Muse est l’agent d’IA personnel de Meta, conçu pour agir de manière proactive sur des objectifs et des tâches, utiliser des applications connectées et le web, et poursuivre des travaux en arrière-plan dans une Muse Secure VM dédiée.", "public_description": "Muse est un agent d’IA personnel présenté par Meta le 8 septembre 2026. Il est conçu pour dépasser l’assistance conversationnelle tour par tour en planifiant et en exécutant des tâches pour le compte de l’utilisateur, en poursuivant des travaux en arrière-plan, en utilisant des services connectés et en revenant vers l’utilisateur lorsqu’une action conséquente exige son approbation.", "technical_description": "Muse fonctionne dans une Muse Secure VM dédiée avec un environnement d’agent isolé, un navigateur, un système de fichiers et un terminal. Sentinel contrôle les permissions des connecteurs et les sorties réseau, tandis que les identifiants sont stockés hors de l’environnement principal de l’agent. Muse utilise la famille de modèles Muse Spark ; la documentation de sécurité publiée au lancement identifie spécifiquement Muse Spark 1.3 pour les workflows de navigation et de résistance aux injections de prompt." } }, "documentation_status": "Connected", "created_at": "2026-09-13T22:37:26+00:00", "updated_at": "2026-09-13T22:37:26+00:00", "reviewed_at": "2026-09-13T22:37:26+00:00", "dypclt_created": "D-0 Y-2 P-4 C-5 L-38 T-2", "dypclt_updated": "D-0 Y-2 P-4 C-5 L-38 T-2", "dypclt_reviewed": "D-0 Y-2 P-4 C-5 L-38 T-2", "temporal_index": { "start_date_utc": "2023-04-01", "created_utc": "2026-09-13T22:37:26+00:00", "created_dypclt": "D-0 Y-2 P-4 C-5 L-38 T-2", "updated_utc": "2026-09-13T22:37:26+00:00", "updated_dypclt": "D-0 Y-2 P-4 C-5 L-38 T-2", "reviewed_utc": "2026-09-13T22:37:26+00:00", "reviewed_dypclt": "D-0 Y-2 P-4 C-5 L-38 T-2" }, "entity_type": "AI Agent", "first_public_appearance_precision": "day", "first_public_appearance_verification": "source_backed_official", "machine_readable_purpose": "Machine-readable registry record for Muse: product identity, agentic capabilities, launch timeline, Muse Spark model lineage, security architecture, official sources, limitations and graph relationships.", "schema_ready": true, "quality_status": "official_source_freshness_review_2026_09_14", "source_policy": "official_first_secondary_context_only", "quality_review": { "official_sources": 7, "trusted_primary_sources": 0, "secondary_sources": 0, "date_status": "source_backed_official", "image_status": "missing", "manual_followup_required": false, "review_scope": "individual official-source review of Muse launch, product design, security architecture and Muse Spark model lineage through 2026-09-14", "freshness_cutoff": "2026-09-14", "freshness_note": "Muse launch on 2026-09-08 and Muse Spark 1.3 launch-era model context are included. Planned future features are explicitly marked as planned." }, "revision": 1, "content_sha256": "03336041bb13fe1a379393c92c1962ba9be1c0cd8e4f0001b7c4b8190b171553" }, "computed": { "category": "Agents", "type": "AI Agent", "subtype": "Personal AI Agent", "documentation_index": { "total": 100, "documentation": 25, "evidence": 25, "structure": 25, "relationships": 25, "level": "Level V Persistent" }, "is_lxkeys_entity": false, "graph_endpoint": "graph.php?center=ENT-00003000&depth=1" } }