chutes-deploy

name	chutes-deploy
description	[BETA — permanent until verified] Deploy models on the Chutes.ai decentralized GPU network from Hermes. vLLM / diffusion / custom CDK deploy, image builds, teeify, rolling updates, stable aliases. Triggers on: deploy chute, chutes deploy, vllm chute, diffusion chute, build chute image, POST /chutes/, teeify, rolling update chutes, model alias deploy.
version	0.1.0
author	Chutes Agent Toolkit
license	MIT
metadata	{"hermes":{"tags":["chutes","deploy","vllm","diffusion","tee","beta"],"status":"beta"}}

Chutes Deploy for Hermes [BETA — permanent until verified live runs]

Status: permanent BETA for deploy-side writes. Wave-2 live verification reached the Chutes deploy API and found server-side gating for easy deploy (HTTP 403 Easy deployment is currently disabled!) on at least some account classes. Scripts now surface fallback hints and resolve branch names to SHAs, but deploy-side writes keep BETA by policy.

When to use this skill

A Hermes user wants to:

• Deploy a new vLLM or diffusion chute from a Hugging Face repo id.
• Build a custom CDK chute image from a Dockerfile + context.
• Teeify an existing affine chute into a TEE-isolated variant.
• Inspect rolling updates on an existing chute.
• Create a stable model alias (interactive-fast, tee-chat, etc.) on top of a deployed chute.

Not for: calling models that already exist on Chutes. That's the hub chutes-ai skill.

Walkthrough (Hermes-facing)

Full walkthroughs and scripts live at plugins/chutes-ai/skills/chutes-deploy/. Hermes users run the same scripts from the repo root:

# Lane A — vLLM
python plugins/chutes-ai/skills/chutes-deploy/scripts/deploy_vllm.py \
  --model Qwen/Qwen3-8B --gpu h100 --alias interactive-fast

# Lane B — diffusion
python plugins/chutes-ai/skills/chutes-deploy/scripts/deploy_diffusion.py \
  --model stabilityai/sdxl-turbo --gpu a100_40gb

# Lane C — custom CDK (two-step)
python plugins/chutes-ai/skills/chutes-deploy/scripts/build_image.py \
  --dockerfile ./Dockerfile --context ./ctx --name myorg/my-chute --tag v1
python plugins/chutes-ai/skills/chutes-deploy/scripts/deploy_custom.py \
  --image-id <id> --entrypoint my_module:chute --gpu h100 --name myorg/my-chute

# Teeify
python plugins/chutes-ai/skills/chutes-deploy/scripts/teeify_chute.py --chute-id <id>

# Alias
python plugins/chutes-ai/skills/chutes-deploy/scripts/alias_deploy.py \
  --alias interactive-fast --model <model_id>

All scripts read cpk_ from the shared keychain via manage_credentials.py.

Deep references

• plugins/chutes-ai/skills/chutes-deploy/SKILL.md — full walkthrough
• plugins/chutes-ai/skills/chutes-deploy/references/vllm-recipe.md
• plugins/chutes-ai/skills/chutes-deploy/references/diffusion-recipe.md
• plugins/chutes-ai/skills/chutes-deploy/references/teeify.md
• plugins/chutes-ai/skills/chutes-deploy/references/rolling-updates.md
• plugins/chutes-ai/skills/chutes-ai/references/model-aliases.md

Safety rules (same as the Claude skill)

• Deploy operations consume real paid compute — always confirm GPU class + cost before POSTing.
• Never hardcode cpk_ anywhere. Always read from manage_credentials.py.
• DELETE, teeify, and make_public are hard to reverse — warn before executing.
• Do not make TEE privacy guarantees without attestation verification. Surface /chutes/{id}/evidence and point at the future chutes-tee skill.

Related Hermes skills

• chutes-ai (Hermes hub) — API key prerequisite.
• chutes-mcp-portability [BETA] — exposes chutes_deploy_vllm, chutes_deploy_diffusion, chutes_teeify, chutes_set_alias as MCP tools (all labeled [BETA]).