Consistent.
A StayCharted-trained model applies a published version to the same inputs, using the categories it learned from your examples.
FOR AI AGENTS · CLAUDE · CHATGPT · API · POWER AUTOMATE
Your agents act. StayCharted makes sure they decide the way your team does.
An AI agent can read a ticket, a claim photo or an invoice. It doesn’t automatically know how your team would handle it. StayCharted is the tool your agent calls for that: a private model that learned from your team’s past decisions. It answers in seconds, with a confidence, and flags the ones it’s unsure about for a person.
Free plan · No code · Your data trains only your models · In the Claude directory
● HOW IT ALL FITS
Your team’s judgment, for every app and agent.
1. Learns from your past work
Support tickets→ Which queue
Leads and messages→ Which reply, who sends it
Invoice lines→ GL account and cost center
Photos and notes→ Pass, or which defect
2. APPLIES IT TO NEW WORK FROM
- AI agents (Claude, ChatGPT)
- Your apps (API)
- Power Automate
- Google Sheets
- Files and photos
StayChartedA private model of your team’s past decisionsAnswers in seconds, with a confidence.✓ Confident: done
Straight into your system, or back to the agent to act on.
Unsure: Review Queue
A person confirms or corrects it, with similar past examples beside it.
Every correction teaches the next version.
A general AI guesses. Your team already knows.
Knows when to ask.
Every answer includes confidence and review status. Your agent can use that signal to hand off instead of acting alone.
Built for volume.
Categorize batches or call the API one item at a time. StayCharted-trained models use plan allowances rather than per-token billing. Connected AI providers bill separately.
The agent asks. StayCharted decides. A person checks the uncertain ones.
- Your agent sends the item: a ticket, a lead, a file, an invoice line, a photo. Through Claude or ChatGPT (AMT's MCP connector), the Prediction API, Power Automate, Google Sheets or a batch file.
- StayCharted answers like your team: the category, or several fields at once, with a confidence.
- Confident answers go straight through. Your integration can route the ticket, suggest invoice codes or file the photo, subject to your permissions and workflow rules.
- Unsure answers go to the Review Queue on Essentials and Business. A person confirms or corrects them, with your team's most similar past examples beside each one.
- Every correction teaches the next version. Reviewers retrain from the queue, and a Builder publishes it, so the next version can learn from reviewed examples.
One queue for every handoff.
On Essentials and Business, the Review Queue is the place where an agent's uncertain decisions wait for a person, whichever channel they came from:
| Where the agent works | What StayCharted decides | Who reviews |
|---|---|---|
| Support agent | Which queue, which reply | A team lead |
| Sales or lead agent | Which reply, which rep | Sales ops |
| Claims agent | Damaged or not, which damage, front or rear | An adjuster |
| Finance agent | GL account and cost center | Accounts payable |
| IT and access agent | Who gets access to a file or SharePoint site, at what level | IT or the data owner |
| Catalog agent | Which category a product goes in | Merchandising |
Every decision has a record.
- A confidence on every answer, and the cutoff you chose for review.
- The model version that answered, with the accuracy it scored on examples it never trained on.
- Activity for every change: who trained, published or reviewed, including changes made through an AI assistant, marked "via Claude".
- A person decides what goes live. Retraining from the queue never publishes on its own.
Your decisions stay yours.
- Your examples train only your workspace's models. Nothing is pooled across customers.
- The assistant sees a model only if you switch it on, for the workspace and for that model.
- Your AI Models: answer with the AI platform your company already uses (Amazon Bedrock, Microsoft Foundry, Google, or an API key), on your own account and bill.
- Need it in your own AWS account? Talk to us about a self-hosted Enterprise deployment.
Works where your agents already run.
Claude
Add StayCharted from the Claude directory.
Read the guide →ChatGPT
Connect StayCharted to use your models from a conversation.
Read the guide →Any agent framework
Call the Prediction API for the label, confidence and review status.
Read the guide →Power Automate
Connect model answers to an approval and application workflow.
Read the guide →Google Sheets
Use the add-on to train and categorize directly in a sheet.
Read the guide →Each Prediction API answer has label, confidence and needsReview. When needsReview is true, your agent can hand off instead of acting. On Essentials and Business, the item is already waiting in the Review Queue.
Review status does not itself enforce your agent’s behavior. Configure the workflow to honor it.
Measured on public data.
93.7% accuracy on the public BANKING77 test set with StayCharted’s Dedicated AI Model. Results depend on the task and data; inspect the method and review tradeoffs before choosing a model.
Read the banking benchmark →Questions about AI agents and human review
Is the Review Queue on every plan?
On Essentials and Business. Every plan says which answers need review.
Is StayCharted an AI agent?
No. It's the decision tool an agent calls. Your agent does the work. StayCharted decides the way your team would, and says when a person should check.
Why not let the agent's own AI decide?
An agent’s general AI does not automatically know your team’s rules and may answer differently across runs. A StayCharted-trained model learns your categories from past decisions, has measured test accuracy and uses plan allowances rather than per-token billing. Your AI Models use your connected provider, which bills separately.
What happens to an answer the model is unsure about?
It is marked for review. On Essentials and Business it also goes to the Review Queue, where a person confirms or corrects it. Corrections can become examples for the next version; a Builder chooses when to publish that version.
Which agents work with StayCharted?
Claude and ChatGPT through AMT's connector, and any agent or workflow that can call an API, including Power Automate. Google Sheets has its own add-on.
Does my data train anyone else's model?
No. Your examples train only your workspace's models.
Can it use our own AI platform?
Yes. Your AI Models answer with your company's own AI platform, such as Amazon Bedrock, on your account. Nothing is trained; they're tested on your held-out examples like every other model.
Explore next
STAYCHARTED AI MODEL TRAINER
Train AI to categorize the way your team does.
Start with examples you already have, see how often it’s right on items it never saw, and use confidence to prioritize the ones a person should review.
Start free