Local Hermes ran a 4-hour code migration on 7M tokens for $0
At a glance
Kai Felix Bennett ran Hermes Agent locally for a four-hour autonomous code migration—7.26M tokens that would have cost about $94 on Claude Opus—controlled from Telegram at zero API spend. This workflow is listed under local-llm, coding-agents, hermes, cost, telegram.
He used Hermes Agent with local llama.cpp inference (Qwen3.6-27B-MTP Q4_K_M on an AMD Radeon AI PRO R9700) and a LiteLLM bridge so Claude Code-style tooling worked against local models. Hermes planned and executed a multi-step code migration, calling tools and editing files while he directed it from Telegram.
How it went
The session processed 7,256,671 tokens over about four hours. Equivalent Claude Opus 4.7 API cost was about $94.34; actual cost was $0.
Save ideas in this browser. Upvotes count across the board.
Runta built ClawShell so OpenClaw/Hermes never hold real LLM API keys—virtual keys map to real ones in a privileged process with DLP scanning and email sender filters.
A Hermes user documents an ANA check-in, baggage handling, airport route alternatives, and a reusable check-in tool, alongside failures in other tasks; the account remains a personal report.
This n8n community template transcribes Telegram voice messages, extracts meal ingredients and estimated nutrients with OpenAI, stores structured rows in Sheets, and replies with a confirmation.
CLI reads Claude Code transcripts to show per-turn context cost—what ships before the first keystroke, unused payload, and skills that never reached the agent—so you can trim defaults instead of guessing.