ToShop DocsHow Tasks Work
What you see in chat when your agent runs a task — thinking, tool cards, results.
A task in ToShop is a chat conversation that may reach for any number of tools along the way. The flow looks the same whether you typed in the app, sent a Telegram message, or a routine fired.
What appears in chat
As your agent works, the chat streams these elements:
Thinking blocks
Expandable sections that show your agent's reasoning. Off by default for fast models; toggle in Settings → Agent.
Tool use cards
One card per tool call. Each shows the tool icon, the inputs (truncated), and a status: running, done, or error.
Skill hint cards
When your agent is about to invoke a skill, you get a short card with the skill name and what it's about to do. You can cancel before it runs.
Message parts
Markdown, code blocks, tables, file attachments, browser snapshots, screenshots, OCR text.
There's no separate "plan / review / finalize" phase you click through. Your agent streams its work, and you intervene when you want — approve, reject, edit the prompt, stop the task.
Approvals during a task
Tool calls that touch your system pause for approval the first time. The card shows the proposed action and offers four options:
Approves just this invocation. Next time the same tool fires with similar args, you'll be asked again.
Approves until you quit ToShop. Useful for batch tasks where you don't want repeated prompts.
Saves a persistent rule in Settings → Permissions. The rule auto-approves matching actions across sessions until you revoke it.
The agent doesn't take the action. You can tell it what to do instead in the same chat turn.
Read-only or pre-approved actions run without prompts.
When your agent keeps going
After a tool call, your agent reads the result and decides what to do next.
If something unexpected comes back, your agent doesn't guess forward — it folds the new information into its next plan. You'll see this as additional reasoning and further tool calls in the same chat.
Stopping a task
The send button becomes a Stop button while the task is running. Stopping interrupts the next step; the tool call already in flight finishes its current operation.
Tasks across surfaces
The same task lifecycle applies whether the trigger was:
In-app chat — visible in real time. The most common entry point.
Connections — runs in the background; your agent posts updates back to the channel thread. See Telegram and Discord.
Routines — runs on a schedule or trigger; results go to the places you configured. See Routines.
Whatever the entry point, your agent has the same skills, memories, and approval rules.
Where the task lives
Conversations are stored on your computer, scoped to each agent. Search across all of an agent's conversations with ⌘K. The audit log of tool calls is at Settings → Local Tools.