Esment for Enterprise

Enterprise is for organizations that need full control over where their AI memory lives. It uses the same engine as the Mac app and the web version, deployed where you decide: on a Mac, on a server, in your network or fully offline. It needs an Enterprise license, which you get by talking to us.

Every memory, search index and knowledge graph entry is stored on hardware you control, in a SQLite database on disk. Nothing is sent to NotasAI, and no cloud account is involved. You can inspect what is stored, where and who can reach it.

• Local models. Embeddings and the reranker run on the machine. The language model that extracts memories from conversations is a local GGUF model served by llama.cpp. See Enterprise models.
• Air-gapped. With the models on disk, no internet connection is needed at runtime.
• Optional outside endpoints. You can point inference at your own OpenAI-compatible or Ollama endpoint instead.

Enterprise uses the same model as the Teams plan, without limits: one account, any number of workspaces, and members with a role in each. A workspace holds its own memories, history, API keys and people, and one team’s key cannot see another team’s memories.

• Roles: Admin, Member and Viewer. Each memory records who saved it.
• Unlimited workspaces, seats and memories on an Enterprise license.
• Enterprise-only features: observational memory (saving from a stream of activity), integrations and the model gateway.

How invitations, roles and seats work is described in Workspaces & members.

Every connection is authenticated. Create API keys per person, tool or integration. A key inherits the permission of the member it belongs to: admin, read write or read only, and can be revoked at any time.

Changes are recorded as commits, so you can diff two versions, walk the audit trail of a single memory or ask what the system believed on a given date.

Enterprise comes in three shapes. Pick the one that fits how your team works:

• The Enterprise Mac app. One signed .dmg that carries the engine and local models. Built for Apple Silicon.
• One binary on your own server. Linux or macOS, with a config file, a port and a health check.
• Docker Compose. The core plus background workers as services, when several teams share one deployment.

A server install is one process with a config.toml. Point your team’s AI tools at the server address, create API keys, and shared memory is available to everyone. Each person connects their own assistant to it.

For licensing, deployment support, security reviews, volume pricing or SLA agreements, get in touch. We help you evaluate Esment and support the rollout.