Control you can see and set.
Canopy agent control sets what each AI agent may do and who approves it. Every workspace runs in one of three send modes: draft only, approve to send or automatic. Every deployment starts on approve to send, each draft shows its evidence and a confidence score, and every agent action is logged as it happens.
Set per workspace, changed any time.
The current level is always visible at the top of each workspace, so nobody has to guess what the agents are allowed to do.
Draft only
Agents research and write. Every message waits in a queue, and nothing leaves until a person sends it. The default for Customers.
Approve to send
Each message arrives as a full preview with its evidence and a confidence score. One click sends it. Every deployment starts here.
Automatic
Agents send, log and learn on their own, and hold any draft under the confidence floor for a person. Earned workspace by workspace.
Deals follow-ups are always approve to send. Open opportunities are too important for any agent to contact on its own, so the rep owns every word.
A preview, not a yes or no.
Quick approve and reject buttons get rubber-stamped at scale. Canopy shows the reviewer exactly what will send and why, so approval means something.
- The exact rendered message, recipient, sender and send time
- The evidence behind every claim, and where each fact came from
- A confidence score, as a percentage and a word: Strong, Good or Look
- Anything under 85% marked Needs a look, so it is read before it goes
In approve to send, every draft waits for a person. In automatic mode, a draft under the confidence floor is held for a person instead of sending.
Every send can be pulled back.
Every approved send waits 30 seconds, and Undo pulls it back. Drafts the agents are unsure of are held for a person, and one switch stops every automatic campaign.
- A 30-second undo on every send
- Drafts under the confidence floor held for a person
- A kill switch that stops every automatic campaign at once
- Dan Ortiz approved Email 1 to Maya Chen
- Lifecycle sent 212 emails in the webinar journey
- Priya Shah edited and approved the Ledgerline follow-up
- GTM Engineer changed Outbound brief to v14, holdout 20%
- Customers paused by Alex Kim for the renewal review
Improves on its own, inside lines you draw.
The self-improvement loop rewrites messaging automatically, but only within limits you set. Here is what changes on its own, and what always waits for a person.
Changes automatically
- Subject lines, openers and calls to action
- Angles and proof points from your approved library
- Send times, sequence order and frequency
- Template choice and journey branching
- Signal weighting and targeting within your ICP
Always waits for approval
- New claims, statistics or customer names
- Offers, discounts and pricing
- New audiences outside your ICP
- Changes to brand voice rules
- Anything in a Deals follow-up before the rep approves
Every change runs against a holdout first. Variants that lose are rolled back automatically, and every promotion is logged with the result that earned it.
How our agents behave.
Show the work
Research appears as a timeline of what was found and why it matters, never as a spinner.
Lead with the decision
Each screen opens on what needs a human today. Data supports the decision.
Ask when unsure
A draft under the confidence floor is held for a person, not guessed at and sent to a customer.
Leave a trail
Who approved what, on which preview, and what the agent knew at the time.
Explore the platform.
Agent control questions.
Anything else, contact us.
What is human-in-the-loop automation?
Automation where AI agents do the work and a person reviews, approves or corrects it at the points that matter. In Canopy, each workspace is set to draft only, approve to send or automatic, every deployment starts on approve to send, and deal follow-ups always wait for the rep.
Where is a workspace's send mode set?
On the Agents page, which lists the send mode for every module and changes it from a menu beside each one. The current mode also shows at the top of each workspace, and Deals follow-ups stay on approve to send.
Can we stop the agents at once?
Yes. The kill switch stops every automatic campaign at once, from the plan or from Slack.
What does the confidence score mean?
It shows how well the evidence supports the draft, as a percentage and a word: Strong, Good or Look. Anything under 85% is marked Needs a look, and a draft under the confidence floor is held for a person even in automatic mode.
How long is the undo window?
30 seconds. After you approve a send, it waits 30 seconds, and Undo pulls it back.
What are the three send modes?
Draft only: agents research and write, and nothing leaves until a person sends it. Approve to send: each message waits as a full preview with its evidence and a confidence score, and one click sends it. Automatic: drafts above the confidence floor send on their own, and anything under it is held for a person.
What does a reviewer see before approving?
The exact rendered message, the recipient, the sender and the send time, plus the evidence behind every claim and where each fact came from, and a confidence score. Anything under 85% is marked Needs a look, so it is read before it goes.
What can the self-improvement loop change without approval?
Within your guardrails: subject lines, openers, calls to action, angles from your approved library, send times, sequence order, frequency, template choice and targeting within your ICP. New claims, offers, pricing, audiences outside your ICP and brand voice changes always wait for a person.
Apply to become a launch partner.
We're onboarding a small group of revenue teams ahead of general availability. Launch partners get founder-led deployment, launch pricing in writing, and a direct line into what we build next.
- Founder-led deployment
- Launch pricing, in writing
- A say in the roadmap