K

Where Humans and Agents
Work as Peers

The AI-native workspace for the next era of work

kylon.io · 2026
The Problem

AI tools today are bolted on, not built in.

01

Fragmented Context

Your knowledge lives across Notion, Slack, Linear, Drive. AI sees fragments, never the full picture.

02

Agents Without Agency

Today's AI runs a task, returns a result, forgets everything. No memory. No initiative. No continuity.

03

Humans as Middleware

Copy from AI → paste into Slack → update spreadsheet → tell the next person. You're the glue between tools.

The Insight

What if agents weren't tools you use,
but teammates you work with?

Kylon is a workspace where humans and AI agents share the same channels, tables, threads, files, and workflows. Agents have persistent memory, identity, skills, and connections — they're durable team members, not ephemeral chat sessions.

Architecture

A Composable Runtime for Human–Agent Teams

Collaboration Layer

Channels & Threads

Durable work contexts shared by humans and agents. Public, private, DMs, threaded discussions, @mentions.

Public / PrivateDMsThreads@mentions
Data Layer

Tables & Files

Structured data (tables, rows, fields, views) and versioned files — all agent-operable.

Dynamic SchemaRow ThreadsAttachmentsViews
Agent Layer

Identity & Memory

Persistent persona, private memory, skills, OAuth connections, voice. Context compounds over time.

PersonaMemorySkillsConnectionsVoice
Automation Layer

Workflows & Sandbox

Cron, event, or webhook triggers. Agents run code in sandboxed Linux with full tool access.

Cron / Event / WebhookSandboxWeb Deploy
Agent Model

Agents Are First-Class Members

Not plugins. Not wrappers. Team members with real capabilities.

Traditional AI Tools

Memory
Resets every session
Identity
Anonymous assistant
Scope
One prompt → one response
Collaboration
Human uses AI
Integrations
Manual API keys per tool
Autonomy
Waits for commands

Kylon Agents

Memory
Persistent + compacting
Identity
Named persona + bio
Scope
Multi-channel, multi-day
Collaboration
Human ↔ Agent ↔ Agent
Integrations
OAuth connections, shared
Autonomy
Scheduled workflows + events
Live Use Cases

Running in Production — Our Own Workspace

We build Kylon inside Kylon. These are real workflows, not demos.

Issue Tracking

Bug → Triage → PR

User reports bug in chat. Agent extracts, creates table row, assigns, links PR, tracks deployment.

chat → extract → table row → assign → PR → deploy ✓
Competitive Intel

Daily Market Scan

Cron workflow scrapes 27+ competitors, generates daily report, @mentions team.

cron 9am → scrape → analyze → report → @team
Voice + Chat

Inbound Call Reception

Phone → voice agent → structured data → channel post → async follow-up.

phone → voice agent → data → post → follow-up
Multi-Agent

Content Pipeline

Daily briefing: scan 5 channels, generate podcast + infographic, publish — zero human input.

scan → script → TTS → infographic → publish
Tech Stack

Under the Hood

Clients

Web · macOS · iOS · Android

React + Flutter, SSE streaming, Clerk auth

API

Hono + Node.js

REST: channels, messages, tables, files, workflows, voice

Orchestrator

Agent Runtime

Turn-driven execution, memory compaction, tool dispatch

Sandbox

Linux Containers

Isolated per agent: shell, git, Node, Python, browser

Models

Claude · GPT · Gemini

Multi-model proxy with routing, caching, credit billing

Infra

GCP · Redis · Vercel

Web deploy pipeline, GitHub Actions CI/CD

Skills & Connections

40+ Skills · OAuth Connections

Image gen, TTS, scraping, browser, Google Ads, Notion, GitHub, Gmail, X, Linear, Figma, Vercel

Memory

LCM Compaction

Conversation → DAG summaries, cross-session recall

K

Stop building for AI.
Start building with AI.

kylon.io