AI Chat
Canopy → AI Chat is the conversational drafting surface. Use it when you don’t have a fixed idea yet and want to think out loud with the model before committing to a draft.
When to use AI Chat
- You’re brainstorming the week’s content angles.
- You want to interview a topic (“ask me 5 questions to find the sharp angle on this”).
- You have a long anecdote and want help finding the LinkedIn-sized story inside it.
- You want to compare two angles side-by-side (“write this as a story vs. as a list”).
For “I have a specific draft and want to polish it” — use Compose instead.
Threads
The left rail shows your saved threads, sorted by last-active. Each thread is a separate context — referring to “the post we drafted earlier” works within a thread, not across.
- New thread — top-left button.
- Archive thread — declutter without deleting.
- Delete thread — destructive; the conversation is gone.
Threads auto-title from your first message.
Composing
Type into the bottom composer and hit Send. The model streams its reply turn by turn. Common patterns:
- “Help me think through what to post this week.” Open-ended. The model asks 2–3 clarifying questions, then proposes angles.
- “Draft 3 different versions of this idea: [paste idea].” Compare-then-pick.
- “Critique this draft as a senior LinkedIn ghostwriter would: [paste].” Skeptical review before publishing.
- “What’s the strongest hook for this draft? [paste].” Targeted fix.
Moving a chat output into a post
Anywhere a chat reply contains a draft post, hover the message and click → Compose. The text moves into the Compose editor as a new draft, where you can refine and schedule.
Drafts created this way are tagged with the thread ID, so you can trace any post back to the conversation that produced it.
What AI Chat won’t do
- Publish. AI Chat doesn’t have access to scheduling or publishing tools. It produces text; you move text into Compose to ship it.
- Read your existing posts. Each thread starts blank. To bring in context (e.g. “match the voice of my last 5 posts”), paste samples into the conversation.
- Hold persistent identity. If you want a “voice” the model matches across sessions, configure tone notes on a Content Agent and use that instead.
Cost
Each user message that gets a response counts as 1 AI credit, regardless of length. Keep a thread open and iterate freely — that’s 1 credit per turn, not per character.
Tips
- Be specific in the first message. “Help me write a post” gets generic. “Help me write a post for SaaS founders about why their first 10 customers should be hand-picked, with a controversial angle” gets sharp.
- Save the system prompt as a thread starter. If you find a prompt formula that works (e.g. “act as a senior LinkedIn ghostwriter, ask me 3 questions, then…”), keep it in a snippet and paste at the start of new threads.
- Iterate, don’t restart. Refine within a thread instead of opening a new one — context compounds.
- Compare angles. “Now write the same idea as a hot take instead of a story” produces useful contrast.