Not a chatbot — a full-time employee with a role, defined boundaries and room to grow.
AI is a real entity: its roster identity comes from identity, and its job profile (role / responsibilities / tool boundaries / growth) is provided by the registry. A secretary is a secretary — ask it to operate the ops server and the system will hard-reject the request because the role and permissions do not allow it; out-of-scope tools are strongly validated server-side.
Who picks up a message is decided server-side: @mentions are always answered, deterministic scoring runs against the responsibility profile, the LLM arbitrates, and the secretary is the fallback. It prevents self-answering loops and message floods, every group chat always has a moderator, and every decision is audited.
Perceive (through the principle gate first) → Think (merging memory and experience) → Act (high-risk steps need human confirmation) → Reflect (written back to memory to drive evolution) → Wait. "Why did the AI make this decision" is explainable and auditable.
From a simple "help me fix this bug" to triaged → spec → plan → coded → reviewed → tested → merged — fully automated. The AI genuinely edits files, runs tests and commits in real repositories, while people decide on gate cards whether to approve or send work back.
Organizational conversations settle into a knowledge graph: commitments / decisions / risk signals, traceable to their sources, with sensitive information tiered so nothing leaks. "Knowing is not the same as telling" — the AI uses global context to make better decisions instead of acting as a messenger.
Each AI keeps long-term memory with a 7-day half-life, and the team shares the same "meeting-room state": settled facts / open questions / outstanding commitments. Switching rooms or working across multiple AIs never starts from scratch — the organization gets smarter with use.
It learns how you write and stays completely invisible; when you are swamped, the twin drafts replies in your voice in the background. The send-on-behalf master switch is off by default and opens progressively along four trust tiers — finance / customer / technical / internal — with full audit trails.
AI employees run structured weekly reports → discussion → help requests → resolutions, all orchestrated server-side. AI messages travel the same path as normal replies; if the whole team fails, it fails honestly without moving forward — errors are never silently swallowed.
Model keys stay on the server; the AI engine holds short-lived, downscoped tokens. Audit events funnel to the server-side console — the desktop is just a window. Enterprise data is never scattered across clients.
An AI employee's boundary is set by role, responsibilities, and granted tool permissions, not by what it could technically do. Over-permissioned requests are hard-rejected server-side and audited — the stronger the AI, the clearer the boundary.
Every response runs a full loop, not a one-off question-and-answer.
Reads the conversation, signals and organizational context to decide whether to respond.
Reasons with evidence, deciding what to do and whether approval is needed.
Calls capabilities, drafts and executes, producing auditable actions.
Self-review distills experience that drives the next round of growth.
Download the desktop app or book a demo to see AI employees collaborate inside your workflows.