Agent
Not a chatbot.
An agent.
Most "AI tools" answer questions. The oxom Agent executes workflows — it knows your workspace, remembers context, and completes multi-step tasks on its own.
The three levels
What's the difference?
Not every AI system is the same. The difference isn't in the model — it's in what the system can do and when it acts.
Chatbot
Responds.
Reacts to questions.
A chatbot understands natural language and gives an answer. That's it. It has no context beyond the current conversation, can't execute anything, has no memory, and doesn't know who you are.
"What's a good title for my next video?"
Gives an answer. Posts nothing, saves nothing, won't remember it next time.
oxom
Agent
Executes.
Plans, acts, completes.
An agent understands the request, plans multiple steps, uses tools — and completes the workflow. It knows your workspace, remembers past decisions, and can execute actions across external systems.
"Shorten the link, generate a QR code, and send both as a Discord announcement."
Runs three tools in sequence. You don't have to click anything.
Autonomous
Acts on its own.
Trigger-based, no input needed.
An autonomous system doesn't wait for input. It responds to events — a new Twitch stream, an incoming sponsor email, a calendar date — and executes workflows without anyone intervening.
Twitch goes live → Agent posts announcement to Discord and Twitter. Automatically.
No prompt. No click. Event-triggered.
How it works
The Agent Loop.
No single request. The agent thinks in loops — until the task is complete. Maximum 8 tool rounds, 25-second timeout per tool.
Prompt
You write or speak a request — or an event triggers the agent.
Planning
The agent analyzes the task and determines which tools are needed and in what order.
Tool Execution
Tools are called — in parallel when possible, sequentially when necessary. Up to 8 rounds.
Context Update
Results flow back into context. The agent decides whether the goal is reached or further steps follow.
Response
The completed workflow is summarized as a response. What was done, what was returned.
Example
One prompt. Three tools. Done.
Prompt
„Shorten the link to my new merch drop, generate a QR code with my logo, and send both as an announcement to the #merch channel."
Tool 1
shortenUrl
Input
https://shop.oxom.de/drop-2024
Output
oxom.co/s/merch24
Tool 2
retrieveQrCode
Input
oxom.site/merch24, Logo: workspace preset
Output
QR code PNG, signed R2 URL
Tool 3
sendDiscordMessage
Input
#merch channel, link + QR code URL
Output
Message ID: 1234, delivered ✓
Response
Link shortened → oxom.site/merch24. QR code generated. Discord announcement sent to #merch.
Tool System
70 Tools. 6 Categories.
Every tool is Zod-validated, workspace-isolated, and configurable per workspace via a permission table. Plus: dynamic MCP servers and CTI metadata per workspace.
Links
Publishing
Analytics
Community
Content
Workspace
Next level
Moving toward Autonomous.
The oxom Agent is currently a reactive agent — it waits for your input. Autonomous means: it acts based on events, without you needing to type anything.
Twitch goes live → announcement on Discord. New sponsor email arrives → classified, prioritized, in the inbox. Calendar event tomorrow → briefing prepared.
The Sponsor Agent on oxom.dev is the first autonomous component — a CF Worker that classifies incoming emails without anyone needing to intervene.
Reactive vs. Autonomous
You write → Agent executes.
Event happens → Agent reacts.
No input needed for routine tasks.
Access
No self-signup.
Curated access.
oxom is designed for 10–40 workspaces. No one-size-fits-all SaaS — every workspace is set up, configured, and supported.