Deploy via Playground

1Open the playground

Visit play.theseus.network. Click your workspace name (top left) → + New workspace → pick BYO Markdown Agent — frontmatter only, no Rust from the dropdown → name it → Create.

You'll see two files in the file tree:

THESEUS.md
skills/
  my-tools/
    SKILL.md

That's the whole agent. No agent.rs.

2Edit THESEUS.md

The template ships as a working “Weather Assistant” you can deploy as-is. To make it yours:

---
name: My Agent                 # display name on chain
id: my-agent-v1                # stable on-chain handle (don't change after deploy)
model: deepseek-chat           # which registered model to use
native-tools: none             # opt out of the built-in tokens.*, chain.*, bridge.* surface
tools:
  - name: get_weather          # MUST match a proxy_tools entry on the tool-executor
    description: Get the current temperature (Celsius) for a city.
    args:
      city: string
    return: string
  - name: lookup_user
    description: Read a user record by id.
    args:
      user_id: string
    return: string
---

You are a helpful assistant.

When the user asks about weather, call `get_weather` with the city.
When the user asks about a user account, call `lookup_user` with the
user id. Otherwise just answer directly.

The body below the frontmatter is the system prompt — same shape and conventions as a Claude or OpenAI system prompt. Be specific about when to call each tool.

Supported arg types: string | number | bigint | boolean.

3Edit skills/my-tools/SKILL.md

The skill is how the model discovers which tools to consider for a task. List each tool's purpose:

---
name: my-tools
description: Catalog of weather + user-lookup tools.
allowed-tools: get_weather lookup_user
---

# Tools

## `get_weather`

Use when the user asks about weather. One argument: `city`. Returns
JSON with `temperature` and `conditions`.

## `lookup_user`

Use when the user asks about a user account. One argument: `user_id`.
Returns `name`, `email`, `preferences`.

The allowed-tools: line must list every tool you want the model to have access to via this skill (space-separated).

4Set up your tools as HTTP endpoints

Your existing tool code keeps running wherever it lives today — you just need to expose each one as a POST endpoint:

POST https://your-server/get_weather
Content-Type: application/json

{"city": "Tokyo"}

…returning the tool's result as the response body. FastAPI, Express, Cloudflare Workers, ngrok during dev — all fine.

Then the operator running the tool-executor needs a proxy_tools: entry per tool in their config:

proxy_tools:
  - name: get_weather
    endpoint: https://your-server/get_weather
  - name: lookup_user
    endpoint: https://your-server/lookup_user

5Deploy

Click Deploy this workspace in the bottom-right. The playground:

1. Reads your THESEUS.md frontmatter
2. Synthesises an agent.rs from the tools: list (you don't see this)
3. Compiles the whole thing via WASM into SCALE bytes
4. Signs with your connected wallet (today: //Alice in dev; wallet integration coming) and submits the register_agent extrinsic

You'll see “Agent registered” with your agent's SS58 and a link to the chain explorer. The agent is live; you can prompt it from the panel right underneath.