Help & support

Getting started

Describe an outcome, not a procedure. “Research the AI video generation market and compare the leading tools” works better than “search for AI video”, because the first one tells Cadive what finished looks like.

Cadive replies in conversation first. When what you asked for is a real piece of work, it offers a task underneath its reply — with the plan, an estimate, and a ceiling. Nothing runs and nothing is charged until you start it.

• Say “run it” and it starts immediately.
• Ask “can you run it as a task?” and it shows you the task and waits.
• Follow up in the same conversation — “make the hero shorter” edits the page it already built, at the same URL.

What a task costs

You pay per task, in cents, from a credit balance. Each step is priced from the tokens it actually used on the model it ran on, so the number you see when a task finishes is the real one — not a rounded guess.

• A short conversational reply is a fraction of a cent.
• A research task that reads several sources and writes a cited report is usually under ten cents.
• Generating and deploying a page is usually a few cents.
• Dictation and read-aloud are billed by the minute of audio, and cost about a twentieth of a cent per minute.

Every task has a hard ceiling and stops there rather than running on: Free $0.25, Pro $5.00, Studio $15.00, Max $40.00. You also see an estimate before anything starts, and the running total while it works.

Modes

Auto reads the task and picks for you. It is the right choice almost always.

• Economy — smaller models, fewer sources. For when you want a quick look.
• Balanced — small models for legwork, the large model for the thinking and the writing.
• Max — larger models throughout, more sources read, longer output.

Tasks default toward output quality rather than finishing fast. If you actually want it quick, say so — “a quick look at…” drops it a mode.

What Cadive can't do

This is the part worth reading, because it is where people expect otherwise.

• It has no access to your accounts. Not your Instagram, not Stripe, not your analytics, not your inbox. Asked to analyse one, it says so rather than inventing figures — and you are not charged for the refusal.
• It can’t see images. You can attach them and they are stored, but nothing reads their contents yet. Describe what matters and it works from that.
• Video is audio only. An uploaded video is transcribed; the visuals are not analysed.
• Some pages block reading. When a source can’t be fetched, the task says which ones failed instead of citing them anyway.

Files

Attach text, code, CSV, audio or video and Cadive works from the contents. CSVs are parsed and their statistics computed in code, not estimated by a model, so the numbers are arithmetic rather than opinion.

• Text and code — up to 40MB, with the first few MB interpreted.
• Audio and video — up to 48MB, transcribed.
• Images — up to 24MB, stored.

Everything you upload and everything Cadive produces is in Files, grouped by what it is.

When something goes wrong

• A task produced less than expected. Sources that couldn’t be read are listed in the task. Ask it to go deeper, or switch to Max.
• It refused. Almost always because the request needs data it cannot reach. It will say what was missing and suggest what it can do instead.
• “You’re out of credit.” Your balance hit zero. Free credit does not renew automatically.
• An edit didn’t apply. Follow-ups edit the work in the same conversation. Starting a new conversation starts fresh.

Contact

Email hello@cadive.net. If it’s about a specific task, include roughly when you ran it and what you asked for — that’s enough to find it.