DEPLOY AGENTS
Deploy like you deploy code.
One-click to API, Slack, Teams, WhatsApp, web, email, voice. Versioned, observable, rollback-ready. Cloud or on-prem. Built for the moment after the eval passes - and the months after that.
·
7 days since deploy
·
Next eval: 2h 14m
Live activity
Real-time event stream
10:42:18
✓
API · POST /agent/refund-specialist · 1.1s
10:42:14
✓
Slack · #cs-internal · escalation triggered
10:42:09
✓
Web · widget session · refund completed
10:42:03
✓
WhatsApp · order lookup
10:41:58
⚠
Email · auto-reply sent · routed to human
Deployed agents handling 10M+ interactions per month.
7+
CHANNELS SUPPORTED
99.97%
UPTIME SLA
< 2s
AVG RESPONSE TIME
1-click
ROLLBACK
CHANNELS
One agent. Every channel your users actually use.
Your users aren't on the channel you'd prefer. They're on Slack at work, WhatsApp at home, email when it's formal, and your website when they're shopping. Deploy the same agent to every one of them - without rebuilding the agent or maintaining six separate integrations.
API / SDK
RESTful API. Native SDKs for Python, Node.js, Go. WebSocket streaming for real-time conversations. Authentication via API keys or OAuth. Rate limits configurable per workspace.
Most common use: custom apps, internal tools, mobile apps.
Webhooks & MCP
Trigger agents from any system that can make an HTTP call. Receive agent outputs at any endpoint. Expose your AgentX agents as MCP servers to other AI tools.
Most common use: workflow orchestration, system-of-record integrations.
Slack
Native Slack app. Mention the agent in a channel, DM it directly, or trigger via slash command. Threads, files, and emoji reactions all supported. Workspace-level installation with admin controls.
Microsoft Teams
Native Teams app with Azure AD SSO. Channel mentions, personal chats, adaptive cards for structured responses. Tenant-level deployment with admin policy compliance.
WhatsApp Business
Production-grade WhatsApp Business API integration. Verified business profile. Rich media support - images, documents, voice notes, location.
Web widget
Embeddable JavaScript widget. White-label friendly. Customizable theme, position, behavior. Mobile-responsive out of the box. Session persistence across page loads.
Dedicated email address per agent. Inbound emails routed to agent, outbound replies sent from your domain via SPF/DKIM. Thread context preserved across replies.
Voice
Phone number per agent. Inbound and outbound calls. Real-time transcription. Tool calls and knowledge access during voice conversations.
VERSIONS
Ship v4. Roll back to v3 in a click.
Every deploy is a versioned snapshot - agent instructions, sub-agent configurations, tool bindings, knowledge sources, model selections, everything. Promote to production when eval passes. Roll back instantly when something looks wrong. No git, no rebuilds, no maintenance window.
🟢 v4 - Production
Deployed 7 days ago by Sarah Kim
Eval 9.4 / 10
142 test cases · 1 override · Changes: RAG retrieval 3→5, added KB "Refund Policy v2"
View diff
Roll back to v3
Promote to v5
🟡 v3 - Previous
Deployed 21 days ago by Marek N.
Eval 8.1 / 10
142 test cases · In production for 14 days · Rolled back
↺ Restore to production
🔵 v2 - Archived
Deployed 45 days ago by Sarah Kim · Eval 7.8 / 10
🔵 v1 - Archived
Initial deployment · Eval 7.2 / 10
✓
✓ Every deploy is a versioned, immutable snapshot.
✓
✓ Rollback completes in seconds - no rebuild, no redeploy pipeline.
✓
✓ Roll back to any prior version, not just the most recent.
✓
✓ Side-by-side diff between any two versions.
OBSERVABILITY
See what your agent is doing. While it's doing it.
A deployed agent is a black box only if the platform makes it one. Every agent action, tool call, knowledge retrieval, sub-agent handoff, and human override is logged, traceable, and searchable. Real-time when you need it, queryable when you don't.
📞 Conversation #4827
✓ resolved · 23.4s
├─ 🤖 Orchestrator - Customer Success Manager - 0.4s
│ ├─ 🧠 Intent classification → "refund_request"
│ └─ ↪ Delegated to: Refund Specialist
│
├─ 🤖 Refund Specialist — 18.2s
│ ├─ 🛠 Tool: Order Lookup → order #ACME-9281 - 1.2s
│ ├─ 📚 KB: Refund Policy v2 → 5 chunks retrieved - 0.6s
│ ├─ 🧠 Decision: amount > $500 → escalate - 0.3s
│ ├─ ↪ Handoff to: Manager Approval Specialist
│ │
│ └─ 🤖 Manager Approval Specialist - 14.1s
│ ├─ 🛠 Tool: Slack notification - 0.4s
│ ├─ 🧑 Human-in-the-loop: approved by mgr.acme - 13.2s
│ └─ 🛠 Tool: Stripe refund - 0.5s - $850.00
│
└─ ✉️ Response sent - 0.8s
Real-time dashboard
Live activity stream. Filter by agent, channel, status, tag. See what's happening as it happens.
Searchable trace history
Every conversation traceable for 90 days (longer on enterprise). Search by user, agent, tool call, error pattern.
Alerts & anomalies
Set thresholds on latency, error rate, escalation rate, score. Slack/email/webhook on threshold breach.
Export to your stack
Stream traces and logs to Datadog, New Relic, Splunk, or any OTel-compatible endpoint.
WHEN THINGS GO WRONG
The interesting question isn't "what if it works" It's "what if it doesn't"
LLM providers have outages. Tools rate-limit. Knowledge bases get out of sync. Edge cases blow up retrieval. Deployed agents on AgentX have explicit fallback rules - so when something fails, the user sees a graceful response, not a stack trace, and your team sees an alert, not a complaint ticket.
Model fallback chain
Primary model unavailable? Fall back to a secondary. Secondary unavailable? Route to a tertiary. Configure the chain per agent. Logged on every fallback event.
Rate limit handling
Tool calls that hit rate limits queue with backoff. Agents that hit per-user rate limits respond with a configurable "I'll get back to you" message and queue the conversation.
Escape hatches
Every agent has a configurable "I don't know how to handle this" path. Route to human queue. Open a ticket. Send to Slack. The user never gets a hallucination as the fallback.
WHERE IT RUNS
Your data doesn't have to leave your environment.
Some agents handle public knowledge and a chatbot persona. Others handle customer PII, regulated financial data, or internal documents that can't leave your perimeter. We run the platform both ways - same UI, same workflows, same eval pipeline, same observability. Different deployment target.
AGENTX CLOUD
☁️ Managed
✓
Managed infrastructure - zero ops on your side
✓
EU data residency available
✓
SOC 2 controls in place (certification in progress)
✓
Default for builders, startups, and most teams
✓
Start free, scale on demand
ON-PREM / VPC
🏠 Self-hosted
✓
Runs in your AWS, Azure, GCP, or bare-metal environment
✓
Data never leaves your perimeter
✓
Air-gapped deployment supported for regulated industries
✓
Compatible with self-hosted LLMs (Llama, Mistral, custom)
✓
Available on Enterprise tier - scoped per deployment
BUILT FOR REAL USERS
Deployment-grade controls. From the free tier up.
Deploying to one Slack channel and deploying to a regulated financial workflow use the same controls. The controls scale with the deployment - they don't change.
Per-deployment RBAC
Different permission scopes per channel, per environment, per workspace. Production deploys require elevated permissions.
Deploy audit trail
Every deploy, rollback, and config change logged. Who, what, when, why. Exportable.
Environment isolation
Test, staging, production environments isolated. Test data never reaches production deployments. Production credentials never reach test environments.
SLA-grade reliability
99.97% uptime SLA on Enterprise. Multi-region failover. Health checks per deployment. Status page per workspace.
You built it. You evaluated it. Now you ship it.
Deployment is the third pillar. The first two are what make this one possible. The eval pipeline gates what gets shipped. The builder defines what gets versioned. Together - Build, Evaluate, Deploy - they're the loop.
Build Agents
Spin up your first agent in the next 10 minutes.
Visual builder, multi-agent orchestration, skills, tools, knowledge, memory. Where the loop starts.
Evaluate Agents
Test cases, LLM-as-judge, versioned eval runs, deploy gate. What earns the deploy button.
GET STARTED
One click from build to production. One more back, if you have to.
Free to start. Deploy to one channel on the free tier, every channel on paid. Cloud by default. On-prem when you need it.
Free
