Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.augustin.ai/llms.txt

Use this file to discover all available pages before exploring further.

Planning doc for the target OpenClaw setup — native OpenClaw agent with all features (memory, skills, heartbeat, cron, browser, identity), using Kiro billing for LLM inference.

​
Goal

A single native OpenClaw agent that has the full OpenClaw feature set (Tyson identity, persistent memory, skills, heartbeat, cron, browser, channel awareness, proactive delivery) while all LLM inference is billed through the Kiro account.

​
Approach

Kiro’s billing is tied to Kiro’s client — there’s no official API to use Kiro credits externally. The path forward is AIClient-2-API, a proxy that reads Kiro’s auth tokens and exposes them as an OpenAI-compatible endpoint. OpenClaw then uses this as a custom model provider, keeping its entire native agent loop intact. OpenClaw’s agent loop handles everything — memory injection, skill loading, tool execution, heartbeat, cron, channel routing. The proxy is just a billing passthrough for the LLM calls.

​
Risk

AIClient-2-API reverse-engineers Kiro’s auth flow. It works today (6.3k stars, actively maintained), but could break if Kiro changes their auth mechanism. Mitigation: if it breaks, swap the provider to a direct Anthropic API key or Bedrock in openclaw.json — one config change, no architectural impact.

​
Chat Frontends

Experimenting with four self-hosted chat UIs, each in its own container, all backed by AIClient-2-API for Kiro billing:

​
OpenClaw Client

Primary: Open WebUI via a Pipe that bridges to OpenClaw’s WebSocket. Polished ChatGPT-like UI, per-conversation context switching, self-hosted. Secondary: Telegram for mobile. First-class OpenClaw channel support, native notifications.

​
Plan

  1. Deploy AIClient-2-API as a sidecar container on i3, pointed at Kiro’s auth tokens ✅
  2. Register it as a custom model provider in OpenClaw’s openclaw.json ✅
  3. Configure the native OpenClaw agent to use it for inference ✅
  4. Verify full feature set works: memory, skills, cron, heartbeat, browser, channel delivery
  5. Deploy Open WebUI with the OpenClaw Pipe
  6. Add Telegram as a channel
  7. Migrate cron jobs (including .kiro/ git sync) into OpenClaw’s cron system