Working in chat
The chat is where you do everything. This page covers the controls, outputs, and costs you'll work with day to day.
Threads: your workspaces
Each conversation is a thread, and each thread keeps its own workspace — a persistent sandbox where files the agent creates stick around between messages. That means you can build something in one message and refine it in the next.
- New chat (left sidebar) starts a fresh thread with a clean workspace.
- Your Recent threads are listed in the sidebar — hover to rename or delete.
- A thread marked Static has had its workspace archived after long inactivity; opening it starts a fresh sandbox.
Sending a task
Type into the box at the bottom ("Ask the agent…") and press Enter (Shift+Enter for a new line). While a task runs, a red Stop button lets you interrupt it.
Attachments: click the paperclip to attach any file — up to 25 MB per file and 100 MB per message. Drag-and-drop and paste-image also work. Attach CSVs to analyze, docs to summarize, images to work from.
Choosing the model
Two controls sit just above the input:
Orchestrator model
A pill shows the model running your orchestrator (a strong general model by default). Click it to open a searchable list of 250+ models, grouped by provider and tagged by tier (cheap / standard / premium / ultra) and capability (web search, code, vision).
Note: the orchestrator model locks once the thread starts. To use a different model, start a new chat.
Default vs. Max
A two-way switch controls how far your sub-agents are allowed to reach:
- Default — sub-agents stay in a low-cost model pool. Cheaper, great for everyday work. New accounts start here.
- Max (⚡) — sub-agents can reach frontier models for the highest quality, at higher cost.
This only affects the sub-agents, not the orchestrator, and every plan and model is available on every subscription — the difference is cost, never access.
Outputs and artifacts
Anything the agent produces for you shows up as an artifact card in the conversation, with a live preview:
- Websites/HTML render in a sandboxed frame
- Images show inline
- Spreadsheets/CSVs preview as a table
- Documents, PDFs, Office files preview in place
- Everything has a Download button
If the agent overwrites a file, you'll see a version badge (v2, v3…). Scratch and working files are tucked into the Files panel (right side) rather than cluttering the chat — open it any time to browse the full workspace.
Sharing a result
Click Share on any artifact card to mint a public link (it's copied to your clipboard). Anyone with the link opens a clean viewer at silico.ai/share/… — no login required — where they can preview and download the file.
A few things to know about shares:
- Sharing is per-file — you share an individual artifact, not the whole conversation.
- The link is public and unlisted — anyone who has it can view, so only share what you're comfortable making public.
- Re-sharing the same file gives you the same link.
Understanding the cost
Under each finished reply, a small receipt shows:
compute 43¢ → charged 51¢ (51 credits · ×4.5) · $2.10 left
That's the real compute cost, what you were charged (real cost × your plan's markup) in credits, and your remaining balance. Each sub-agent card also shows its own cost. If a run would be expensive, Silico can pause to show an estimate before spending. See Plans & credits for the full picture.