What are Bots?
Publish an agent once. Call it from your application or an AI app.
A Bot is an agent you configure and publish in Stacklink. Give it instructions, a model and the tools it may use. Stacklink runs the model, calls permitted tools and saves the result.
Build into your app
TypeScript and Python. Run a Bot, get its answer and continue the conversation.
Use from an AI app
Ask your connected AI app to find and run a published Bot through MCP.
One Bot, two ways to use it
const run = await stacklink.bots.run(botId, {
sessionId: session.id,
message: "Summarize these meeting notes into three next steps.",
});
const result = await stacklink.bots.wait(botId, run.id);
if (result.status === "completed") console.log(result.finalText);run = stacklink.bots.run(
bot_id,
session_id=session.id,
message="Summarize these meeting notes into three next steps.",
)
result = stacklink.bots.wait(bot_id, run["id"])
if result["status"] == "completed":
print(result["finalText"])Use Stacklink's published "Meeting notes" bot in my project.
Ask it to summarize the notes below into three next steps.
Show me the Bot's result.
[paste your notes]These are usage previews. Start with the complete SDK example or MCP walkthrough.
Create your first Bot
Create
Open Bots in your Stacklink project and create a Bot. Start with one clear job, such as turning meeting notes into an action list.
Configure
Choose a model and write its instructions. Add only the capabilities it needs.
Try it
Send a message in the builder. Sample mode avoids live external calls; use Live deliberately when testing connected tools.
Publish
Enable Your app for SDK calls or AI apps for MCP. Publishing saves the configuration callers will use. Later draft edits do not change that published configuration until you publish again.
Give it useful work
| Bot | Add what it needs |
|---|---|
| Meeting notes | Instructions; structured output if your app needs fields |
| Order assistant | Your API for authorized order data; optional connected support tools |
| Scheduling assistant | A supported scheduling/calendar connection and its permitted actions |
| Research assistant | Web search, page reading and selected knowledge sources |
| Operations assistant | Connected apps, approval permissions and optional Workbench |
Who runs the model?
With Bots, Stacklink runs the agent and produces the answer. If you already run your own agent loop, use Stacklink's tool SDK integrations instead.
Know what is available
Bots support multi-turn conversations, past-chat search, approvals, schedules and event-triggered runs. Your application supplies the customer-facing chat UI.
Slack and Teams conversational deployments use a separate setup from SDK/MCP access. See channel setup. Each deployment needs the Bot’s own installed app identity before it can be activated.
Durable questions with reminders, human handoff, Telegram and WhatsApp conversations remain outside this release. Approval notifications in team apps are a separate feature.