Chat

Hold persistent conversations with general-purpose AI agents, attach files, switch models and watch agent activity stream in real time.

What chat is for

Chat is the conversational surface of the platform. Unlike a pentest — which is a structured engagement against a defined scope — a chat is an open-ended session with a general-purpose agent. Use it to reason about a target, draft remediation guidance, interpret a finding, or ask an agent to run a one-off investigation with its tools.

Chats are organised around the platform’s general agents (phase “General”, distinct from the pentest phases). They keep full history, so you can pick up a conversation days later with all of its context intact.

Start a conversation

Sessions and persistence

Every conversation is a persistent session you can return to:

• Your chats live under My chats — sessions you created and own.
• Shared chats appear under Shared chats — conversations a teammate has shared into a team you belong to.

Each message and tool call is stored as an operation in the session, so the full transcript is preserved. You can:

• Rename a chat to keep your list organised.
• Archive a chat to move it out of the active list without deleting it, and unarchive it later.
• Share a chat with a team so colleagues can read and continue it, or unshare it from one team or all teams.

Sharing is how chat becomes collaborative: a triage discussion or an investigation can be handed to the whole team rather than living in one person’s account. Sharing requires a team — see Teams & RBAC.

Selecting agents and models

Two independent choices shape a conversation:

• Agent — a general-purpose agent with its own system prompt, assigned tools and (optionally) MCP servers. Different agents are tuned for different jobs.
• Model — the underlying LLM the agent runs on. The models available to you depend on your tier; higher tiers unlock more and larger models. Your assigned models are shown in the model picker.

If you have built or cloned your own agents, they appear here alongside the platform defaults. To create and customise agents — assign tools, attach MCP servers, swap the model — see Agents, tools & MCP.

Attachments

You can attach files to a message when the selected model supports file input (the picker indicates which models accept files). Attachments are sent along with your prompt so the agent can read them — for example a configuration file, a log excerpt or a screenshot to analyse. Models that don’t support files simply won’t offer the attach option.

Real-time agent activity

When an agent works, you don’t just wait for a final answer — you watch it think and act. The platform streams a live timeline of agent events over Server-Sent Events from the streaming backend. As the agent runs, you’ll see:

• Thinking — the model’s reasoning as it decides what to do next.
• Tool calls and results — each tool the agent invokes, its arguments, and the result, with status icons for normal, cached, skipped or blocked calls.
• Interpretation — a short “analysing results…” phase after tools run, with the analysis text streaming in token by token.
• Iteration progress — the current iteration out of the agent’s maximum, plus running findings counts.
• Subagents — when an agent spawns helpers, they appear nested beneath it, each with its own mini-timeline.

The result is a transparent view of the agent’s work rather than a black box. The same event stream powers the live view during pentests; if you want to consume it programmatically, see Streaming.

Where to go next