Code Inference — Development skill for Claude Code
Local inference stack for coding agents.
How to install Code Inference
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Code-Inference/code-inference and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Code Inference does
Local inference stack for coding agents. Run opencode, Claude Code, Cursor, Codex or Grok Build against a preconfigured llama.cpp gateway — no cloud, no prompt leaves your host.
Alternatives in Development
- Llamacoder — Open source Claude Artifacts – built with Llama 3.1 405B 7.1k ★
- Agentic Stack — One brain, many harnesses 2.2k ★
- Ship Studio — The professional studio for modern development — build it, ship it, host it without leaving the app 249 ★
README
code-inference
Local inference stack for coding agents — runs opencode, Claude Code, Cursor, Codex or Grok Build in your project against a preconfigured local `llama.cpp` stack.
Quick install
# Latest (pre-release)
curl -sS https://raw.githubusercontent.com/Code-Inference/code-inference/development/install.sh | sh
# Stable release (once on main):
# curl -sS https://raw.githubusercontent.com/Code-Inference/code-inference/main/install.sh | sh
Usage
From any project directory:
code-inference # launch opencode with the inference stack
code-inference --fresh # standalone opencode (no inference, persistent volumes)
code-inference --restart # restart the inference stack (volumes preserved)
code-inference --full-isolation # compose with scoped project name (inference, ephemeral)
code-inference --full-isolation --disk-name EXT1TB # per-project volumes on an external disk
code-inference --help # show usage
Choosing an agent
`opencode` is the default. Pass `--agent` **as the first argument** to use another one:
code-inference --agent claude # Claude Code
code-inference --agent cursor # Cursor CLI
code-inference --agent codex # Codex CLI
code-inference --agent grok # Grok Build
code-inference --agent claude --fresh # agent alone, no inference stack
code-inference --agent codex --restart # restart that agent's stack
`--agent` after a mode flag is an error, not a silent fallback:
code-inference --fresh --agent claude # Error: --agent must be the first argument.
Every agent gets its own user, home and volumes, so credentials, session history and settings never collide with opencode's or with each other's. See [docs/agents.md](docs/agents.md) for the per-agent reference: versions, base images, volume layout, project templates, and how to add another agent.
`--full-isolation --privileged --disk-name
Related Skills
Q Agent
Self-hosted llama.cpp stack for Qwen + MTP (multi-token prediction): a 125B MoE read from ≥ 80 GB of RAM, or a
Learn Grok Bot
Learn Grok Bot's desktop agent architecture in 16 lessons: Electron process isolation, trusted RPC, turn loop,
Cpp Stop Ledger
Reconcile every STOP #1 plan against the evidence. Regenerates the derived disposition ledger — which audits a
Libertai CLI
LibertAI CLI — single-binary tool for inference, image generation, and launching agent tools (Claude Code, Ope
Agent Proxy
Linux API gateway for Claude Code, Codex, Google Antigravity, and Grok CLI tools
SDK Crossplatform Eval
PHẢI dùng khi cần đánh giá khả năng thay 1 SDK thiết bị native/Windows-only đang dùng (P/Invoke, DLL/.so cũ) b
Related Agents
Model Runner
Deterministic wrapper that runs a prompt on a non-Claude model (OpenAI via Codex, Cursor's composer/grok/glm,
Cursor Delegate
Proactively use when this coding agent should hand a substantial coding, diagnosis, or implementation task to
Plugin Host Parity Checker
Checks that the shared skills/ tree, the three host manifests (.claude-plugin/, .cursor-plugin/, .codex-plugin