📌 Roadmap

This tracks the development status of Capsule.

v0.1.0: The Python Runner

Status: ✅ Done

Goal: Validating the core Host/Guest technology and Python interoperability.

• Core: Rust Host capable of loading Wasm Components.
• SDK (Python): Basic @task decorator with JSON serialization.
• CLI: capsule run main.py with JIT compilation.
• Limits: Basic Fuel metering for CPU protection.
• Resilience: Retry logic and real Timeout handling (Wall clock).

v0.2.0: Production & Polyglot

Status: ✅ Done

Goal: Stability, security, and expanding the ecosystem.

• TypeScript SDK: Functional wrapper support (const t = task(...)).
• Better Logs: Better logging for debugging and monitoring.
• Distribution: CI/CD for binaries and Package Registries (PyPI, npm, crates.io).

v0.3.0: Data Access (Filesystem)

Status: ✅ Done

Goal: Enable agents to work with local files and datasets.

• Filesystem: Local file mounting (fs_access) for reading images, CSVs, and datasets.

v0.4.0: Configuration & Observability

Status: ✅ Done

Goal: Improve developer workflow with project configuration and richer task feedback.

• Config File: capsule.toml for default task options.
• Structured Output: Tasks return a detailed JSON envelope with execution metadata:

{
  "success": true,
  "result": "<task return value>",
  "error": null,
  "execution": {
    "task_name": "analyze_data",
    "duration_ms": 1523,
    "retries": 0,
    "fuel_consumed": 45000
  }
}

On failure:

{
  "success": false,
  "result": null,
  "error": {
    "type": "timeout",
    "message": "Task exceeded 30s timeout limit"
  },
  "execution": {
    "task_name": "analyze_data",
    "duration_ms": 30000,
    "retries": 1,
    "fuel_consumed": null
  }
}

v0.5.0: Node.js Ecosystem Compatibility Extended

Status: ✅ Done

Goal: Enable natural Node.js development patterns and improve npm package compatibility in WASM.

Polyfills needed:

• path: Use path-browserify and alias import path from 'path'.
• os: Custom minimal polyfill for import os from 'os'.
• process: Comprehensive process polyfill (env, argv, cwd, exit, nextTick) injected as a global via esbuild.
• url: Alias Node's import { URL } from 'url' to native Web URL class.
• buffer: Use buffer package to polyfill Node.js Buffer class → Uint8Array.
• events: Use events package to polyfill EventEmitter.
• stream: Use readable-stream for Node.js stream compatibility.
• stream-web: Use stream-web for Node.js stream/web compatibility.
• fs: Map import fs from 'fs' to WASI file operations:
  • fs.readFile() → Capsule's readText()/readBytes()
  • fs.writeFile() → Capsule's writeText()/writeBytes()
  • fs.readdir() → Capsule's list()

v0.6.0: SDK Runner & Network Controls

Status: ✅ Done

Goal: Enable seamless third-party integration and basic network security.

• SDK Runner: Add run() function to both SDKs for async subprocess execution.

// JavaScript/TypeScript
import { run } from '@capsule-run/sdk/runner';

const result = await run({
  file: './capsule.ts',
  args: ['code to execute']
});

# Python
from capsule import run

result = await run(
    file="./capsule.py",
    args=["code to execute"]
)

• Allowed Hosts: Restrict outgoing HTTP requests to a whitelist of domains.

// Allowed Hosts
export const main = task({
  name: "main",
  allowedHosts: ["api.openai.com", "*.anthropic.com"],
}, () => { ... });

v0.7.0: Prebuilt Sandbox Adapters & Network Support

Status: ✅ Done

Goal: Allow to easily use prebuilt sandboxed environments with ready‑to‑use adapters.

• Python Adapter: Prebuilt sandbox template for Python frameworks.
• TypeScript Adapter: Prebuilt sandbox template for TypeScript/JavaScript frameworks.
• Socket shim for Python: Transparent support for standard network libraries (urllib) via a custom socket implementation (bypasses componentize-py limitation)
• Enable filesystem mode: read-only and read-write to extend possibilities

v0.8.0: Richer execution metrics

Status: 🚧 In progress

Goal: Provide more visibility into each task's behavior for better auditing and monitoring.

• Network : Return which domains/IPs were called and what data was sent
• RAM: Return the exact amount of memory consumed by the task
• Filesystem: Return which files were read, created or modified (useful when --mount is used)
• Environment: Return which environment variables were accessed