What Is Kylon AI?
Pure Reason's agent workspace, explained in plain language.
Kylon is a shared team workspace where named artificial intelligence (AI) agents with durable memory and scoped permissions work in dedicated rooms alongside your human staff. Pure Reason makes the software, describing it as a company harness for multi-agent work rather than a standalone chat assistant or document layer.
At Layer3Labs, we build and run AI systems inside other people's businesses, and the hardest part of agent deployment is rarely getting a model to answer. Most of the work goes into sharing context between agents, revoking tool credentials safely, and stopping several agents from overwriting what people are doing.
Kylon launched publicly in October 2026 for web, macOS, iOS, and Android. It lets teams create specialized internal agents or connect external coding tools such as Claude Code and OpenAI Codex directly into company threads.
What Kylon Is and How It Works
In practice, you add an agent to a room the way you add a colleague. It keeps reviewable memory across conversations and acts only through connections you authorize and can revoke. Agents reply in the room's threads, run saved workflows on a schedule or trigger, and send consequential actions such as emails back to a person as drafts to approve. You can also invite Claude Code, Codex or Devin Cloud into a room as teammates.
Workspaces start with a default head of agents that runs onboarding.
- Maker: Pure Reason (co-founded by Quinn Leng).
- Category: Multi-agent company harness and team workspace.
- Supported platforms: Web, macOS, iOS, and Android.
- Base pricing: Core plan starts from $50 per month for 500 monthly credits.
- Trial terms: 7-day free trial with 300 credits and up to 5 agents, credit card required.
- Security status: General Data Protection Regulation (GDPR) compliant, with SOC 2 Type II certification in progress.
How Kylon Rooms and Agent Permissions Work
Rooms are Kylon's shared channels, where people and agents work in the same threads. MindStudio reports that each room carries its own apps, automations, files and knowledge base.
Agent memory is reviewable context an agent keeps across conversations, so you can check what an agent has retained instead of rereading old threads. Kylon's own line is "A transcript is not memory."
Permissions are set per agent, per user and per room, according to Kylon's Kimi Work comparison. Each connection is scoped by permission and can be revoked, so one agent's access can be withdrawn without changing anyone else's.
Consequential actions come back to a person to approve. When an agent writes an email or a social post, Kylon saves it as a draft, and a person sends it.
A newly brought-in agent starts private rather than open to the whole workspace, and it has its own Admin and Member roles.
How to Bring Claude Code, Codex or Devin Into Kylon
Kylon supports external coding agents, allowing tools such as Claude Code, OpenAI Codex, and Devin Cloud to join workspace channels.
If you have not used these agents before, our guides to Claude Code and OpenAI Codex explain how each one works in the terminal. Devin connects differently, through a credential form that uses Cognition's cloud API. Gemini CLI, Grok Build, Hermes and OpenClaw work only in Kylon deployments that enable them.
External agents connected through your local terminal rely on your hardware. If your computer is turned off or falls asleep, that self-connected agent goes offline and reconnects automatically once the machine wakes up. If the team depends on that agent, run it on a machine that stays awake.
Usage for these self-connected agents is billed directly by your own model provider rather than through Kylon credits, so heavy coding work does not drain the workspace's credit balance. The underlying Kylon CLI also supports developer tooling including Cursor, Antigravity, Amp, Droid, Kiro, Cline, OpenCode, and GitHub Copilot, as documented on Kylon's guide to connecting agents and tools.
- 1. In Kylon, open Members.
- 2. Choose "Invite my agent". Kylon gives you a prompt and an invite link that works once and expires after 30 minutes.
- 3. Paste the prompt into Claude Code, Codex or another terminal agent.
- 4. The agent installs the Kylon CLI and a background service on macOS, Linux or Windows, then joins the workspace as a teammate.
Kylon Workflows and Chat-Built Apps
Workflows are saved tasks that run on a schedule or a trigger, and each keeps a run history so you can see what ran and when. Pure Reason says a workflow invokes an agent that reasons through the task instead of following fixed steps.
Skills are packages of domain knowledge for agents. Pure Reason has stated it plans to open-source its skills and templates.
You can build a database app, such as a CRM or an inventory tracker, by describing it to an agent in chat.
These custom applications run on serverless Postgres hosted by Neon, store their underlying code in Git repositories, and deploy on Vercel. Agents render outputs as interactive cards, tables, charts, per-user views, and approval buttons directly inside room feeds.
Kylon Integrations and MCP Connectors
Kylon advertises access to more than 3,000 software integrations across every plan, with the majority delivered through Composio. Kylon builds native connectors for Gmail, Google Docs, Google Sheets, GitHub, Notion, and X.
Linear, Supabase, Neon and Plaud connect to Kylon using each vendor's official Model Context Protocol (MCP) server.
Kylon also embeds access to built-in services billed directly against your Kylon credit balance without requiring independent vendor accounts. These embedded services include Apollo, Ahrefs, Apify, Exa, ElevenLabs, GPT Image, DataForSEO, and OpenRouter.
Standard language model calls to Anthropic's Claude, OpenAI's GPT, and Google's Gemini route through Kylon's internal proxy service.
Kylon Pricing and Credit Allowances
Kylon calculates subscription fees per workspace rather than charging per user seat. As of October 2026, the entry-level Core tier starts from $50 per month, granting 500 base credits, up to 10 active agents, email support, and a recurring 10 percent bonus credit allocation.
Higher usage tiers within the Core subscription scale linearly: $100 per month delivers 1,000 credits, $200 delivers 2,000, $400 delivers 4,000, $600 delivers 6,000, and $800 delivers 8,000 credits. Across all Core tiers, every dollar spent purchases exactly 10 base credits, and the 10 percent bonus provides 550 usable monthly credits on the minimum $50 plan.
Teams that need more than 10 agents or extra credit packs move to Pro or Ultra, which are priced by quote ("Talk to sales").
Pure Reason labels its Bring Your Own Agent (BYOA) tier a pricing concept whose amounts and eligibility are not final. Monthly credits reset each billing cycle, and agents and workflows pause when credits run out. The full tier table is in our Kylon pricing breakdown.
Who Kylon Fits and Where It Falls Short
Kylon fits operational, marketing, and sales departments that coordinate multi-step campaigns across multiple external tools. Common workflows include pulling analytics from Meta Ads and Google Analytics, running prospecting searches through Apollo, and staging draft emails for account executives.
A common snag in agent rollouts is credentials: each person wires their own API keys into their own agent, and nobody can revoke access in one place. Kylon's per-agent connections, which an admin can revoke, address that directly.
Kylon does not publish how many credits a task uses, so run your real workflows during the trial and watch the credit balance before you pick a tier.
Security-sensitive enterprises should note that Kylon holds GDPR compliance but lists its SOC 2 Type II certification as in progress, without published HIPAA or ISO 27001 attestations. If your security review needs a finished SOC 2 Type II report, ask Kylon for its timeline; the current status is on the what is Kylon page. There is no free plan, and self-connected agents go offline when the host computer sleeps.
If your team works in Slack and wants agents there instead of in a separate workspace, compare Kylon vs Claude Tag. For a wider view of where agents fit, see autonomous AI agents for business.
How to Test Kylon During the 7-Day Trial
New teams can evaluate Kylon by registering for its 7-day free trial on the web. The trial requires a credit card and provides full Core plan functionality with 300 trial credits and up to 5 agents.
Set up one room for a single job your team repeats, such as the weekly marketing report or outbound drafts. Connect only the tools that job needs, for example Gmail and Google Sheets, and schedule one or two workflows so they run several times within the 7 days.
Track how quickly your 300 trial credits deplete during these test tasks. Unused trial credits expire immediately when the seven days end, so measuring real usage rates will reveal which Core credit tier your team actually requires before paid billing starts.
Frequently Asked Questions
- Kylon AI is a shared workspace made by Pure Reason where human team members and AI agents work together in dedicated rooms. Each agent maintains its own role, persistent memory, and scoped tool permissions.
- Kylon is made by Pure Reason. Pure Reason's blog names Quinn Leng as a co-founder, and Kylon launched publicly in October 2026.
- Kylon does not replace Claude Code or OpenAI Codex. It connects them into the workspace via its CLI so entire teams can prompt and interact with those coding agents in shared rooms.
- Kylon does not offer a permanent free plan. It provides a 7-day free trial with 300 credits and up to 5 agents, after which Core plans start from $50 per month.
- Kylon is GDPR compliant, and its SOC 2 Type II certification is currently in progress. Kylon encrypts data in transit and at rest, but it does not claim HIPAA or ISO 27001 certification.
- Teams can build scheduled or triggered workflows, packaged skills and database applications such as internal CRMs and inventory trackers by chatting with an agent. Apps deploy to Vercel and run on serverless Postgres via Neon.
Rolling Out AI Agents to Your Team?
Layer3Labs helps teams choose an agent workspace, wire it into their tools and set permissions and spend limits before agents touch real work.
Book a Consultation