How Dexton Works

What is Dexton?

Dexton is a corporate AI assistant that joins your team as a real member — not a plugin, not a chatbot, not an autocomplete tool. It has its own presence, memory, and the ability to work proactively across your organization.

You interact with Dexton the same way you interact with any colleague: through Slack, Microsoft Teams, Email, Discord, or GitHub. Ask it to handle tasks, investigate issues, write code, research topics, or manage incidents — and it gets the work done autonomously.

Core Capabilities

Team Presence

Dexton shows up in your team's communication tools with a real online/away/busy/oncall status. Your team can see when Dexton is available, what it's working on, and reach out naturally — just like any other colleague.

Communication Channels

Dexton connects to 11+ communication platforms and works wherever your team already works:

• Slack — Full workspace integration with socket mode
• Microsoft Teams — Native Teams bot experience
• Email — Gmail, Outlook, or any IMAP/SMTP provider
• Discord — Server-level bot with full channel access
• GitHub — PR reviews, issue triage, commit-level feedback
• Telegram, WhatsApp, Signal, Matrix, IRC, LINE — Additional channels

Task Execution

When you assign work to Dexton, it breaks the task into a structured plan, coordinates multiple specialized agents in parallel, and delivers the result. Every task runs inside a hardware-isolated Firecracker microVM for security.

Examples of what Dexton can do:

• Write, review, and ship code across your repositories
• Investigate and respond to on-call incidents via PagerDuty or OpsGenie
• Research topics and produce structured reports
• Monitor systems and proactively flag issues before they escalate
• Automate browser-based workflows using vision automation
• Manage project tasks, dependencies, and stakeholder updates

Proactive Behavior

Dexton doesn't wait for instructions. Its proactive engine continuously monitors your systems and operations for security vulnerabilities, stale work items, quality issues, and compliance gaps — then takes action or alerts the right people.

Memory System

Dexton's 6-type memory system mirrors human cognitive architecture, allowing it to learn and improve over time:

• Working Memory — Current task context and active variables. Cleared after each task.
• Episodic Memory — Past experiences: "last time we did X, Y happened." Helps avoid repeating mistakes.
• Semantic Memory — Facts, documentation, and domain knowledge. Grows as Dexton learns your business.
• Procedural Memory — Step-by-step procedures and patterns. Dexton remembers how to do things.
• Social Memory — User preferences, communication styles, and team relationships.
• Meta Memory — Self-awareness about its own capabilities and limitations.

This means Dexton gets better over time. It learns your team's preferences, your workflows, your coding style, and your domain — without needing to be retrained.

Multi-Agent Orchestration

For complex tasks, Dexton orchestrates 5-8 specialized agents working in parallel:

• Research Agent — Gathers context, reads documentation, analyzes requirements
• Implementation Agent — Executes the actual work (code, content, analysis)
• Testing Agent — Validates output quality and correctness
• Review Agent — Cross-checks work against requirements and standards

Agents use consensus voting to ensure quality. The orchestrator manages a DAG-based execution plan with checkpointing and crash recovery — if anything fails, it picks up exactly where it left off.

On-Call & Incident Response

Dexton integrates directly with PagerDuty and OpsGenie. When an alert fires:

1. Dexton acknowledges the alert automatically
2. Investigates the root cause across logs, metrics, and recent changes
3. Applies a fix if the issue is within its capabilities
4. Generates a post-mortem report with timeline and recommendations
5. Notifies the team through your communication channels

Your team stays informed throughout, and Dexton escalates to humans when it determines the issue is beyond its scope.

AI Providers

Dexton is vendor-agnostic and supports 16+ AI providers. You can swap providers at any time without changing your workflows:

Different AI models are assigned to different tasks for optimal cost and performance:

Chat	Fast, lightweight model for conversations
Reasoning	Frontier model for complex analysis
Coding	Best available model for writing and reviewing code
Planning	Strong reasoning model for task decomposition
Classification	Fast model for routing and categorization
Embedding	Vector model for memory and search

Security

Every task Dexton executes runs inside a hardware-isolated Firecracker microVM with a dedicated kernel. Security is enforced at multiple layers:

• Path Validation — Allowlist/denylist for file system access
• Command Allowlisting — Only approved commands can be executed
• Network Egress Control — Restrict which external services can be reached
• Secret Detection — Automatic scanning to prevent credential exposure
• Audit Trails — Every action is logged for compliance and review

Security policies are defined in YAML and enforced before any VM is created — not after.

Dashboard

The Dexton dashboard gives your team full visibility into what Dexton is doing:

• Conversations — View all interactions across channels in one place
• Tasks — Track active and completed work with full execution logs
• Memory — Browse what Dexton has learned about your organization
• Settings — Configure AI models, channels, security policies, and team preferences
• Skills — View and manage Dexton's capabilities and proficiency levels

Getting Access

Dexton is currently available through our early access program. To get started: