John johnandersen777

Cocore function calling — enable on local machine

Reproducible guide. Tested on macOS + Apple Silicon M5 + cocore v0.9.50. Also covers headless Pi setup.

What this gets you

OpenAI-compatible function/tool calling through cocore cloud, routed to your local inference provider. Model emits structured JSON tool_calls. Requester executes. Nothing runs on inference host.

droplet-oidc-poc demo

$ mkdir rbac
cd rbac
git init
Initialized empty Git repository in /home/johnandersen777/.tmp/tmp.pVjmEYIeok/rbac/.git/
$ vim setup.sh
$ bash -xe !$
bash -xe setup.sh

Claude Code + Local Gemma 4 12B (llama.cpp / ROCm)

Run Claude Code against a local google/gemma-4-12B-it-qat-q4_0-gguf model via llama.cpp in Docker with ROCm GPU acceleration, fronted by a litellm proxy.

$ uvx whichllm@latest
      Built dbgpu==2025.12
Installed 25 packages in 20ms

╭───────────────────────────── Hardware Info ────────────────────────────────────────────────────╮

gemma-4-12B-it-qat-q4_0-gguf:Q4_0 local on RX 9070 XT

Opencode

{
  "$schema": "https://opencode.ai/config.json",
  "model": "llama.cpp/gemma-4-12b",
  "provider": {
    "llama.cpp": {

Compute Contract Flow

End-to-end walkthrough of the RFP → Bid → Accept → Receipt lifecycle as expressed by the lexicons under lexicons/com/publicdomainrelay/temp/. Records are shown in YAML for readability; on the wire they are JSON records living in ATProto repositories.

Actors

| Actor | Role |

Compute Contract Flow

End-to-end walkthrough of the RFP → Bid → Accept → Receipt lifecycle as expressed by the lexicons under lexicons/com/publicdomainrelay/temp/. Records are shown in YAML for readability; on the wire they are JSON records living in ATProto repositories.

Actors

| Actor | Role |

Qwen3.6-35B-A3B-MTP local on RX 9070 XT

ggml-org/llama.cpp#22673

What worked

docker/model-runner:mtp (image 7b6f81c6dc4b) has the MTP-patched llama.cpp baked in (FROM llama-rocm:full). Retag it as :latest because docker model status/run auto-pull and clobber :latest:
```
docker tag docker/model-runner:mtp docker/model-runner:latest
```

	#!/usr/bin/env bash

	export ANTHROPIC_MODEL="deepseek-v4-pro[1m]"
	export ANTHROPIC_DEFAULT_OPUS_MODEL="deepseek-v4-pro[1m]"
	export ANTHROPIC_DEFAULT_SONNET_MODEL="deepseek-v4-pro[1m]"
	export ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-v4-flash"
	export CLAUDE_CODE_SUBAGENT_MODEL="deepseek-v4-flash"
	export CLAUDE_CODE_EFFORT_LEVEL="high"
	export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
	export ANTHROPIC_API_KEY="sk-..."

	#!/usr/bin/env bash
	# Claude Code -> Charm Hyper (GLM-5.2) via a local litellm proxy.
	#
	# Hyper's Anthropic /v1/messages endpoint rejects `{"role":"system"}` inside
	# the messages array (400 unsupported message role: system), which Claude Code
	# emits. Hyper's OpenAI /v1/chat/completions endpoint accepts it. litellm
	# translates Anthropic Messages -> OpenAI Chat Completions, moving system into
	# the OpenAI messages array. So we run litellm locally on an ephemeral port,
	# point the Claude CLI at it, and tear it down when claude exits.

	This is free and unencumbered software released into the public domain.
	Anyone is free to copy, modify, publish, use, compile, sell, or
	distribute this software, either in source code form or as a compiled
	binary, for any purpose, commercial or non-commercial, and by any
	means.

	In jurisdictions that recognize copyright laws, the author or authors
	of this software dedicate any and all copyright interest in the
	software to the public domain. We make this dedication for the benefit
	of the public at large and to the detriment of our heirs and