Agent harness token overhead: where it goes
Two harnesses on one model can bill an order of magnitude apart. The mechanics behind the gap, and how to measure your own agent's token spend.
Filtering by topic #agent-harness · clear
Two harnesses on one model can bill an order of magnitude apart. The mechanics behind the gap, and how to measure your own agent's token spend.
Run Codex, Claude Code and Hermes behind one self-hosted API. The exact Docker deploy, the loopback bind, the default login you must change, and TLS access.
An agent harness is the program around the model: the loop, the tools, the permissions and the session state. How it differs from a model and a framework.
Build a dsh plugin from an empty folder: the package.json fields that matter, the patch file that mounts it, a real tool, and the two hooks you need.
DeepSeek Harness, Claude Code and Omnigent compared on architecture, model coupling, licence and maturity, plus what each one takes to run on a VPS.
Run dsh, the DeepSeek Harness, as a systemd service on a VPS: dedicated user, pinned version, Restart rules, journalctl logs, and an SSH tunnel to the UI.
dsh prints http://127.0.0.1:3080 because the Web UI binds to localhost only. Reach it safely with an SSH tunnel, and see why publishing 3080 is unsafe.
Installing a dsh plugin runs someone else's code with your agent's permissions. What a plugin can reach, and how to check one before you install it.
Every published DeepSeek Harness build is a release candidate. Pin an exact dsh version, clear the npx cache, and check which npm your Node bundles.
Where dsh keeps its config on Linux, how to wire a DeepSeek API key or a local Ollama endpoint, and exactly what leaves your box in each mode.
Install the DeepSeek Harness on a Linux VPS, pin the npm version, understand what a plugin can do, and reach the port 3080 web UI over an SSH tunnel.