[agenticwork]
← blog

AI Agents in Your Channels: Slack and Teams Integration

Your team already lives in Slack and Microsoft Teams. That is where decisions are made, incidents are triaged, and status updates are shared. What if your AI agents lived there too — not as novelty chatbots, but as full participants in your operational workflow?

Our platform now integrates natively with both Slack and Microsoft Teams, bringing the full power of AgenticWorkflows, SmartModelRouter, and the rest of the platform directly into your messaging channels. This is not a webhook wrapper or a bot framework. It is your deployed platform infrastructure, accessible from the tools your team already uses.

What This Actually Looks Like

The integration supports three interaction patterns, each serving a different operational need.

1. Conversational AI via Direct Message

Send a message to the platform bot in any channel or DM. The message is routed through your deployed platform instance — through SmartModelRouter, through your configured model policies, through your audit trail — and the response comes back to the channel. This is not a thin wrapper around a model API. Every interaction passes through your full security stack: credential isolation, DLP scanning, RBAC checks, and immutable audit logging.

The experience is conversational. Ask a question, get an answer. Ask a follow-up, get context-aware continuation. But unlike consumer AI chatbots, every token is routed through your infrastructure, governed by your policies, and logged in your audit system.

2. Trigger AgenticWorkflows from Messages

Use a slash command or @mention to kick off full AgenticWorkflows flows from a channel. These are not simple request-response interactions. They are multi-step, multi-tool autonomous workflows that can run for minutes or hours, accessing your connected systems, making decisions, and reporting results back to the channel as they execute.

Workflow Trigger Flow
Slack / Teams
Webhook Gateway
Platform API
Flow Executor
SmartModelRouter
Response

Examples of workflows you can trigger from a message:

  • “Run a security scan on the staging environment” — kicks off a multi-step flow that enumerates resources, checks configurations, runs vulnerability assessments, and posts a summary report back to the channel.
  • “Generate a customer onboarding report for Q1” — queries your connected systems, aggregates data, generates a formatted report, and drops it in the thread.
  • “Deploy the latest build to staging” — triggers a deployment pipeline flow with human-in-the-loop approval gates, posting status updates as each stage completes.

3. Persistent Channel Agents

The most powerful pattern: persistent flow agents that run on schedules or in response to events, continuously monitoring and reporting in designated channels. These agents are not triggered by user messages — they run autonomously, posting updates, alerts, and reports as events occur.

Example: Always-On Cost Reporting

A persistent channel agent can be configured to post GCP cost summaries to a Slack channel every morning. The flow runs on a schedule, queries the billing API, compares against budget thresholds, and posts a formatted summary with day-over-day trends, top cost drivers, and projected month-end spend.

Cost Bot8:00 AM
Daily GCP Cost Report — March 10, 2026
Yesterday
$4.82
Month-to-Date
$47.30
Top services:
Cloud Run........... $2.14 ↑12%
Artifact Registry.... $1.08
Cloud DNS........... $0.62
Cloud Logging....... $0.51
Projected month-end: $148.50 (within $200 budget)

This is not a static dashboard you need to remember to check. It is a live report that appears in the channel your team is already watching. If costs spike unexpectedly, the agent flags it immediately. If a new service appears in the billing data, the agent calls it out. The team sees it where they see everything else — in the channel.

Example: Autonomous Incident Response

When a monitoring alert fires — a PagerDuty webhook, a Prometheus alert, a Cloud Monitoring notification — an AgenticWorkflows flow spins up automatically. The flow investigates the issue: checks service health, examines recent deployments, queries logs, correlates with known patterns. Then it posts its findings and recommended actions to the ops channel.

Autonomous Incident Response Flow
1Alert fires
PagerDuty / Prometheus / CloudWatch
2Flow activates
AgenticWorkflows picks up the event
3Investigation
Checks health, logs, recent deploys
4Analysis
Correlates with known patterns
5Posts findings
Summary + recommendations to #ops
6Awaits approval
HITL gate before remediation

The agent does not remediate automatically by default — that is configurable, and most teams prefer a human-in-the-loop gate before automated remediation. But it is designed to perform the initial investigation work that often consumes significant on-call engineer time, and to post its analysis to the channel shortly after the alert fires. By the time the engineer opens the channel, the context is already there.

Security: Your Infrastructure, Your Data

A critical distinction between the platform’s integration and consumer AI chatbots: every response comes from your authenticated platform deployment. Not from a third-party bot service. Not from a shared multi-tenant API. From your infrastructure, running your model policies, using your credentials.

Credential Isolation
Each integration uses scoped OAuth tokens. Your Slack bot token never touches model providers. Model API keys never appear in Slack.
Full Audit Trail
Every message, every workflow trigger, every model invocation is logged with user identity, timestamp, and full request/response payload.
DLP Scanning
Responses are scanned for PII, credentials, and sensitive data before they reach the channel. Configurable redaction policies per channel.

This matters because Slack and Teams channels often include sensitive operational discussions. An AI agent that responds in those channels needs the same security posture as any other system that handles sensitive data. Consumer AI integrations — where messages are sent to a third-party API, processed on shared infrastructure, and returned through a generic bot — do not meet this bar.

How It Compares

The Slack/Teams integration market has a lot of noise. Here is how the platform’s approach differs from the common alternatives.

Our Platform
Generic Slack Bots
ChatGPT for Teams
Model choice
Any provider via SmartModelRouter
Fixed to one API
OpenAI only
Workflow engine
Full AgenticWorkflows
Webhook callbacks
None
Audit trail
Immutable, compliance-grade
Basic logging
Microsoft logs
Data residency
Your infrastructure
Third-party servers
Microsoft cloud
Credential handling
Isolated per-user OBO auth
Shared bot tokens
Microsoft SSO
HITL gates
Configurable per workflow
Manual only
None
Custom tools
OAT tool synthesis
Static webhook endpoints
GPT Actions

Architecture

The integration is architecturally simple, which is deliberate. Slack and Teams both support webhook-based event delivery. When a message is sent to the platform bot or a slash command is invoked, the messaging platform sends a webhook to the Platform Gateway — which is already running in your K8s cluster as part of the standard deployment.

Integration Architecture
Slack / Teams──webhook──▶Platform Gateway
Auth + RBAC check
Flow Executor / Chat Handler
SmartModelRouter──▶Model Providers
DLP Scan + Audit Log
Slack / Teams◀──response──Platform Gateway

The Gateway handles authentication, routes the request to either the chat handler (for conversational interactions) or the flow executor (for workflow triggers), passes through SmartModelRouter for model selection, runs DLP scanning on the response, logs everything to the audit system, and returns the response to the channel. The entire round-trip happens within your infrastructure.

For persistent channel agents, the flow executor runs on a configurable schedule (cron-based) and posts results to designated channels via the Slack/Teams API. The agent does not need to be “triggered” — it runs autonomously and reports results as they are generated.

Getting Started

The Slack and Teams integrations are available to all platform deployments. Setup involves creating a Slack App or Teams App with the appropriate webhook permissions, configuring the webhook URL to point to your Platform Gateway, and defining which channels the integration should operate in.

For persistent channel agents, you define the flow in AgenticWorkflows and configure the schedule and target channel. The flow executor handles the rest.

If you are running the platform and want to add messaging integration, the configuration is straightforward. If you are evaluating AI platforms and messaging integration is a requirement, this is a capability that most platforms either do not offer or offer only through third-party bot frameworks that introduce additional security and operational complexity.

Resources