costeffective-coding-with-local

Overview¶

Everything in costeffective-coding, plus local model backends (Qwen models via Ollama-compatible servers). Routes cloud and local inference through the same LiteLLM proxy.

cf s i -pa costeffective-coding-with-local

What's Added¶

This recipe extends costeffective-coding with:

Local provider — two Qwen models on separate ports
Local secrets — LOCAL_M_API_KEY and LOCAL_S_API_KEY
Custom alias — custom maps to Qwen3.6-27B instead of DeepSeek-V4-Flash
Image router — Qwen3.6-35B-A3B added to text-only image routing

Local Provider¶

Routes to any OpenAI-compatible inference server running on the host machine. Two pre-configured models on separate ports.

Variable	Description
`LOCAL_M_BASE_URL`	Primary model endpoint (default: port 8000)
`LOCAL_M_API_KEY`	Primary model key (any non-empty value)
`LOCAL_S_BASE_URL`	Secondary model endpoint (default: port 8001)
`LOCAL_S_API_KEY`	Secondary model key (any non-empty value)

Models:

Model	ID	Context	Output	Notes
Qwen3.6-27B	`openai/qwen3.6_27b`	131K	16K	Reasoning enabled, seed 42
Qwen3.6-35B-A3B	`openai/qwen3.6_35b_a3b`	262K	41K	Vision-capable, reasoning enabled

Local model parameters:

Parameter	Value
Temperature	0.0
Top-p	1.0
Top-k	1
Max thinking tokens	1,536
Seed	42
Repetition penalty	1.0
Thinking	Enabled (`enable_thinking: true`)

Local models use the system-message-merger plugin to handle system messages correctly.

Changed Secrets¶

Secret	Description
`LOCAL_M_API_KEY`	Primary local model key (any non-empty value)
`LOCAL_S_API_KEY`	Secondary local model key (any non-empty value)

All costeffective-coding secrets are also required:

Secret	Description
`LITELLM_MASTER_KEY`	Proxy auth key
`MICROSOFT_FOUNDRY_API_KEY`	Azure Foundry API key
`OPENCODE_ZEN_API_KEY`	OpenCode API key
`OPENROUTER_API_KEY`	OpenRouter API key
`GITHUB_PERSONAL_ACCESS_TOKEN`	GitHub token for MCP

Set up all secrets:

bash ~/.codefreedom/scripts/costeffective-coding-with-local/setup-secrets.sh

Changed Model Aliases¶

Alias	costeffective-coding	costeffective-coding-with-local
`fable`	Qwen3.7-Max	Qwen3.7-Max
`opus`	Qwen3.7-Plus	Qwen3.7-Plus
`sonnet`	DeepSeek-V4-Pro	DeepSeek-V4-Pro
`haiku`	DeepSeek-V4-Flash	DeepSeek-V4-Flash
`custom`	DeepSeek-V4-Flash	Qwen3.6-27B

The custom alias points to the local Qwen3.6-27B model instead of cloud.

Providers (Inherited)¶

All providers from costeffective-coding are included:

Azure Foundry — GPT-5.x models
OpenCode Zen — MiMo, DeepSeek, Qwen, MiniMax, Kimi (free + subscription)
OpenRouter — DeepSeek, MiMo, Qwen, MiniMax, Kimi, FreeRouter

Agent Profiles¶

Identical to costeffective-coding. See costeffective-coding profiles.

Tools¶

Inherited from _default. See costeffective-coding tools.

Files Deployed¶

~/.codefreedom/
├── .env.claude.secrets          # Claude Code secrets
├── .env.mimo.secrets            # MiMo Code secrets
├── .env.opencode.secrets        # OpenCode secrets
├── .env.proxy.secrets           # Proxy provider secrets (includes LOCAL_*)
├── profiles/
│   ├── claude-code.yaml         # Claude Code agent profiles
│   ├── mimo-code.yaml           # MiMo Code agent profiles
│   ├── open-code.yaml           # OpenCode agent profiles
│   ├── chrome.yaml              # Chrome tool config
│   ├── web.yaml                 # Web search tool config
│   ├── github.yaml              # GitHub tool config
│   └── web-bridge.yaml          # Web bridge tool config
├── proxy/
│   ├── docker-compose.yaml      # Container orchestration
│   └── config/
│       ├── config.yaml          # LiteLLM main config (includes local.yaml)
│       ├── providers/
│       │   ├── azure-foundry.yaml
│       │   ├── opencode.yaml
│       │   ├── openrouter.yaml
│       │   └── local.yaml       # ← NEW: local Qwen models
│       └── plugins/
│           └── reasoning-efforts/
│               └── reasoning-efforts-mapping.yaml
└── scripts/
    └── costeffective-coding-with-local/
        ├── setup-secrets.sh
        └── setup-secrets.ps1

Differences from costeffective-coding¶

Aspect	costeffective-coding	costeffective-coding-with-local
Providers	Azure, OpenCode, OpenRouter	Azure, OpenCode, OpenRouter, Local
`custom` alias	DeepSeek-V4-Flash	Qwen3.6-27B
Local secrets	—	`LOCAL_M_API_KEY`, `LOCAL_S_API_KEY`
Image router models	FreeRouter	FreeRouter, Qwen3.6-35B-A3B
`config.yaml` includes	3 providers	4 providers (+ `local.yaml`)
Setup script path	`scripts/costeffective-coding/`	`scripts/costeffective-coding-with-local/`