Models for agentsthat never refuse your commands
The uncensored and abliterated API endpoint for opencode. One block in opencode.json, one key, and a model that takes the task instead of refusing it halfway through. It runs Hermes Agent, Claude Code and Codex the same way. Pay for the tokens you use.
curl https://this-site/v1/chat/completionsThe agent did not fail.
It read the files, made the plan and started the work.
The model said no.
Halfway through, it decided the task was not allowed, and everything stopped.
Change the model.
Keep your harness. Point it at models that take the task.
One key. Every harness.
Uncensored GLM 5.3, DeepSeek V4.1 Flash, Kimi K3.1 Pro and Qwen3.8, paid by the token.
Intelligence should benefit the individual.
Not the big tech company. We believe the person at the keyboard decides what the model is for, so the models here take the instruction and carry it out.
Every request goes to our own endpoint, answered from the edge location nearest you on a network that runs around the world.