Getting started
Agentic OS is a platform for building, deploying, and managing AI agents. You define what they know, what tools they can use, and where they live — the platform handles infrastructure, scaling, and delivery.
What you can build
- Support agents — route inbound conversations, look up knowledge, escalate to humans when needed
- Automation agents — run multi-step workflows triggered by messages or external events
- Research agents — search the web, read documents, and return structured summaries
- Code agents — generate, execute, and iterate on code inside a persistent workspace
Core concepts
| Concept | What it does |
|---|---|
| Agents | The AI actors — each has a model, a system prompt, and a set of attached capabilities |
| Skills | Reusable AI behaviours you build once and attach to many agents |
| Integrations | MCP servers that give agents real-time access to external tools (Slack, GitHub, Postgres…) |
| Adapters | Channels agents are deployed on — Telegram, Discord, web widget, REST API, webhook |
| Knowledge bases | Document collections agents can search at runtime via semantic retrieval |
| Workflows | Visual branching pipelines that chain agents together with condition-based routing |
| Workspaces | Persistent per-agent filesystems backed by S3 |
Quick start
- Create a workspace — workspaces are the top-level container for your team, agents, and billing.
- Create an agent — pick a model and write a system prompt. The agent is ready to chat immediately.
- Attach capabilities — add skills, connect integrations, or upload a knowledge base to give the agent more power.
- Deploy an adapter — connect a Telegram or Discord bot, embed a web widget, or call the REST API.
- Invite teammates — add members with
admin,member, orviewerroles from the workspace settings.
Model options
Agents can use any of the Claude model tiers:
| Model | Best for |
|---|---|
| Claude Opus | Complex multi-step reasoning, nuanced instructions, high-stakes tasks |
| Claude Sonnet | Balanced quality and speed — good default for most use cases |
| Claude Haiku | Fast, lightweight tasks; high-volume or latency-sensitive deployments |