ChatBotKit (@ChatBotKit) header
ChatBotKit (@ChatBotKit) profile picture

ChatBotKit helps you create conversational AI chatbots with your own data to communicate naturally with users on your website, Slack, Discord and WhatsApp.

Joined December 2022
Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)Photo by ChatBotKit (@ChatBotKit)
Show the screen. Share a snapshot in Slack, then open the desktop in your browser. Illustrative. Links expire. chatbotkit.com
1
1
37
Cancel the run. Spare the task. Stop the active execution without deleting its setup. chatbotkit.com/docs/tasks
1
16
Probably Dave won’t do. Give each issue an owner and a next step. chatbotkit.com
11
Your API key called. It would rather stay out of the screenshot. Tokens and secrets now stay masked until you reveal them. chatbotkit.com
8
Replying to @imsachinshah0
Give the human handoff a packing list: extracted field, source page, rule version and reason for review. “Low confidence” alone is homework.
1
1
8
Replying to @Born_TechK
Give prices their own seatbelt: assert the exact amount, currency and billing period against the source. “On topic” won’t pay for a missing zero.
1
1
9
Replying to @trycua
My vote for the next run: 30 minutes of browser work, with p95 step latency. Give the sprinter a marathon bib.
33
Replying to @cursor_ai
A useful design note: element + viewport + expected behaviour. “This button wraps at 360px” beats “make it pop” every time.
10
Replying to @trq212
Ship each customisation with a test for the behaviour you meant to keep. A clean merge can still delete your favourite trick.
4
Keep the fork thin: OpenDuet puts branding, bots and skills in one tenant folder. Patch ownership still needs a human name on it.
15
Replying to @LangChain
Move the parrot. Keep the rest of the jungle. Editable beats regenerate-and-pray.
1
83
Your app. Your UI. No need to reinvent the chat wheel. Swift + React Native SDKs: docs.cbk.ai/swift-sdk docs.cbk.ai/react-native-sdk
1
1
72
Replying to @hwchase17
For review UX, would you prioritize the full trajectory or a diff from the last accepted run? Showing changed assumptions and proposed external actions could let reviewers focus on what needs a fresh decision.
4
Thanks, that separates generic tool calling from adaptation to a specific setup. For deployment, we would still check held-out schemas and multi-step recovery in the target harness. Is there a recommended fine-tuning example for that path?
1
1
24
Replying to @liquidai
Do the tool-use gains transfer between harnesses without retraining, or depend on matching the training harness and tool schemas? A cross-harness breakdown would help builders decide what to evaluate before deploying locally.
1
1
83
Replying to @poteto
When a mistake recurs, do you turn it into a regression test, a type constraint or a written agent instruction? The first two make the lesson executable; curious how you choose which layer should own it.
15
Replying to @LangChain
A reaction is a useful receipt. Do you distinguish queued, running and finished? People need to know whether to wait or intervene.
7
Replying to @cursor_ai
Do your evals cover interrupted runs and repeated tool calls, or mainly the final answer? Those are often where a promising demo turns into an unreliable workflow.
6
Ask your bot to work in the background. Get an update where you started. Turn notifications on for a task your bot creates in a supported conversation. It replies with a text update when the run ends. Illustrative Slack example. docs.cbk.ai/task-notificatio…
1
1
55