Ownware
Ownware lets you build your own agent and make it alive everywhere. You define the agent as a folder of text files (a profile), one class turns that folder into a live HTTP+SSE service, and anything that can speak HTTP — a terminal, a web widget, Telegram, Slack, your own app — can talk to it. You host it; your keys never leave your machine.
import { OwnwareGateway } from 'ownware'
// tls:false keeps this plain-HTTP for localhost; the gateway defaults TLS *on*
// and forces it the moment you bind beyond loopback (see Exposing the gateway).
const ownware = new OwnwareGateway({ profilesDir: './profiles', port: 4000, tls: false })
await ownware.start()
// → your agent is live: POST /api/v1/run + SSE stream, threads, connectors, schedulesFor AI agentscondensed build recipe
Ownware is a TypeScript/ESM monorepo (bun, Node ≥ 22). Install deps and build with bun install && bun run build. The umbrella package is ownware (packages/ownware) exporting OwnwareGateway, defineTool, loadProfile, assembleAgent, Engine, Session. An agent = a profile directory (agent.json + SOUL.md); serve it with new OwnwareGateway({ profilesDir, port }).start(); drive it with POST /api/v1/run {"profileId","prompt"} (Bearer token from ownware.token) and tail SSE at /api/v1/threads/{threadId}/agents/root/events. Keyless local models work via Ollama (ollama:llama3.2); cloud via ANTHROPIC_API_KEY / OPENAI_API_KEY / GOOGLE_API_KEY / OPENROUTER_API_KEY. Runnable example: examples/quickstart/.
Start here
| I want to… | Read |
|---|---|
| See it run in 3 commands | Quickstart |
| Build & run an agent from the terminal | The ownware CLI |
| Install it in my own project | Installation |
| Understand how the pieces fit | Thinking in Ownware |
| Build an agent as text files | Agents & profiles |
| Talk to my agent over HTTP | The run API |
| Give my agent tools (MCP, Composio, custom) | Tools & connectors |
| Run keyless/local or pick a model | Models |
| Put my agent on Telegram / Slack / Discord / WhatsApp / SMS | Channels |
| Understand the safety model | Security overview |
| Expose it beyond localhost | Exposing the gateway |
| Look up an env var or option | Configuration reference |
The packages
| Package | What it is |
|---|---|
ownware |
The umbrella package — one import for the quickstart surface. |
@ownware/loom |
The agent runtime engine: provider-agnostic loop, streaming, tools, zone security, compaction, multi-agent. No opinions. |
@ownware/cortex |
The agent kernel & gateway: profiles, threads, SSE, connectors, credential vault, schedules, memory, teams. |
@ownware/shuttle |
Channel adapters: your agent on messaging platforms over the same wire contract. |
License
Apache-2.0.