Chat
The conversational surface for talking to your agents directly in the dashboard.
Chat is where you talk to your agents directly. It's the fastest way to use the platform: pick an agent, send a message, and it responds in chat runtime mode with its full identity, memory, and assigned tools.
Resources it contains and manages
| Resource | What it manages |
|---|---|
| Chat sessions | Named conversations scoped to an agent or council (optionally a project). Tracks title, last activity, message count, token estimates, and channel (chat vs notifications) |
| Messages | Individual encrypted turns (user / assistant). Each session records its history for context, review, and later retrieval by agents via memory |
Using chat
- Choose an agent to set who you're talking to. Each agent brings its own prompt, memory profile, and capability surface. The current agent is shown in the header (as
@name) and the composer placeholder hints at the active context. - Messages are private. Your messages and the agent's responses are encrypted on your device before they reach the platform — the server only stores ciphertext and routing metadata. See Security.
Special input commands
The composer recognizes three special prefixes that open pickers, switch context, or invoke behaviors. These are handled client-side before the message is sent and work on both desktop and mobile.
| Prefix | What it does | How it works |
|---|---|---|
@ | Agent switch / @-mention navigation | Typing @ (or mid-message) opens a filtered agent picker. Selecting an agent advances to a session picker showing recent chats with that agent plus a "+ New chat" option (optionally scoped to a project or "General"). This lets you fluidly jump between agents and sessions without leaving the chat view or manually creating sessions elsewhere. The flow supports keyboard navigation and filtering; leftover @text is cleaned up on completion or escape. |
# | Domain activation | When the current agent has assigned domains, typing # shows a domain picker (by command e.g. #creator and display name). Selecting a domain (or typing #command [message]) creates a domain session and sends the (optional) follow-on message under that mode. The domain's developer prompt addon is appended, extra abilities/scopes may apply, and the activation is recorded for audit. While a domain is active, special exit commands (#exit, #done, #finish) or the UI exit the mode cleanly. Domains can also be pre-selected via pill before typing the message. |
/ | Slash commands | Typing / opens the command palette. Always includes the built-in /new-session ("Create a fresh chat with this agent and project."). Installed packages contribute additional commands (skills/commands/hooks adapted from Claude plugins or native packages). Selecting a package command usually sets a pending pill in the composer; type any arguments and send—the full invocation (e.g. /my-command args) is sent together with commandInvocation metadata. Direct typing of a known command also matches and invokes. /new-session bypasses the normal send path. |
These prefixes make chat a high-bandwidth control surface for ad-hoc work, context switching, and package capabilities. See the Domains and Agents primitives for the underlying models, and Packages for how commands are contributed.
Chat is best for exploration, quick questions, and driving Nenji to inspect or build your workspace. For repeatable, multi-step work, reach for a routine.
Dashboard
The main workspace overview — stat cards for projects, agents, running work and tokens, the interactive token usage panel, and recent activity feed.
Agents
The dashboard hub for building agents and the behavioral resources around them — domains, abilities, context blocks, MCP servers, and models.