orka.Orky takes the first pass.
You decide how far it goes.
Draft a reply, answer a familiar question, summarize a conversation. Bring your own knowledge, keep a person close, and make the cost understandable.
Meet your new teammate
Orky suggests an answer in the composer. Read it, change it and send it as yourself. A suggested reply is not a message that has already gone to the visitor.
Hi Jessy! You can update the booking from your account. I can help you find it.
Send as you ↗Let Orky answer from your website, help documents and FAQs. Enable it where you want assistance and take over when the conversation needs a person.
The context stays with the conversation.
Get help saying hello or goodbye, drafting an answer, making a little conversation or summarizing a thread. Two-way translation helps people write in their own language.
Publish accurate articles once. People can search them in your help center, read them in your widget, and get assistance informed by the same knowledge. Improve the source when the product changes.
Explore the help center ↗Start with the number of conversations you expect. Here, one conversation means 10 customer messages and 10 AI replies.
Example: 60 tokens per customer message, 120 per reply, plus 1,500 tokens of instructions and knowledge per reply. The growing conversation history is read again.
About $0.017 per conversation in this example
23.7M input × $0.50/M + 1.2M output × $4.00/M
A planning estimate, not an included allowance or a fixed per-conversation price. Team plan and taxes are separate.In the everyday-support example, one complete conversation uses about 23,700 input tokens + 1,200 output tokens. One million input tokens covers roughly 42 conversations, which also need about 50,400 output tokens. One million output tokens covers roughly 833 conversations, which also need about 19.74 million input tokens. These are two parts of the same bill, not interchangeable allowances.
We assume 10 customer messages and 10 AI replies, 60 tokens per customer message, 120 per reply, and 1,500 tokens of instructions and knowledge on each reply. Input includes the growing history: 10 × 1,500 + 55 × 60 + 45 × 120 = 23,700. Output is 10 × 120 = 1,200. Message length, retrieved articles, summaries, translation and additional model calls change actual usage. These are examples, not measurements of your customers.
The estimate uses uncached input and generated output. It excludes caching discounts, additional reasoning tokens if used, and tool charges. Only provider usage records determine the final bill.
| Model | Provider input / output | Orka input / output |
|---|---|---|
| GPT-5 mini ↗ | $0.25 / $2.00 | $0.50 / $4.00 |
| Claude Haiku 4.5 ↗ | $1.00 / $5.00 | $2.00 / $10.00 |
Your instructions, supplied knowledge, the new message and earlier messages all contribute. Reading the history again uses tokens again.
A short confirmation and a detailed troubleshooting answer use different amounts. Ten replies do not have a universal token count.
Managed usage uses standard provider rates ×2, with input and output counted separately. Bring your own key to pay your provider directly, without an Orka markup.
API rates, not ChatGPT or Claude consumer subscriptions. Provider-specific charges use the same ×2 multiplier for managed usage. Free-plan conversation translation requires your own key. Check the model and current rate before enabling paid usage.
AI can get things wrong. Start with reviewed drafts, keep your knowledge current and use a human for exceptions. The current product catalogue does not promise autonomous refunds, legal advice or every possible back-office action. A clear answer and an authorized action are different jobs.
You choose when Orky participates in a conversation. Fleet also lets you give Orky a custom name. The goal is a teammate that fits your way of helping, while people remain responsible for the commitments made to customers.
See the complete feature catalogue ↗YOUR NEXT CONVERSATION STARTS HERELive chat. Shared knowledge. An AI teammate.
And a little room for the rest of your life.