Skip to main content
Glama
stevederico

dottie-local

by stevederico
OverviewSchemaSourceRelated ServersDiscussionsAdmin
JavaScript
Local

dottie-local

Local llama.cpp inference + optional dotbot harness.

Replaces local-ai-cli for LLM (ask + llm-server). Same CLI names on PATH after install. STT/TTS stays dottie-talk (transcribe / speak).

Same product shape as dottie-talk: HTTP + MCP + CLI. Talk is voice; this is tokens.

MIT. Public.

What it is

Layer

Role

llama-server

Engine (GGUF → tokens). HTTP only.

dottie-local

Façade: start/health, OpenAI proxy, MCP, CLI

dotbot

Optional agent loop + dot_* tools on top

Two processes. Never vendors Metal into the harness.

Related MCP server: LM Studio MCP Bridge

Requirements

  • Node.js ≥ 22

  • llama-server on PATH

  • A GGUF (or let -hf download). Default model: ggml-org/gemma-4-12B-it-GGUF (prefers cached Q8_0)

Install

git clone https://github.com/stevederico/dottie-local.git
cd dottie-local
npm install
npm link          # puts ask, llm-server, dottie-local on PATH

Remove or unlink old local-ai-cli ask / llm-server first so PATH does not shadow.

Quick start

ask "hello"                                    # drop-in for local-ai-cli
echo "$(cat article.txt)" | ask "summarize this"
LLM_REASON=1 ask "prove it"

llm-server start|stop|restart|status|log

dottie-local agent "What tools do you have?"   # harness + dot_* tools
dottie-local start                             # HTTP :1318
dottie-local health

Use with dottie-desktop

Point Settings Provider at Dottie Local (or set localBaseUrl). Base URL is the façade — not raw :8080:

http://127.0.0.1:1318 — needs llama-server on PATH + dottie-local start. Model list: GET /api/tags. Chat: POST /v1/chat/completions.

Ports

Port / env

Service

8080 (LLM_PORT / DOTTIE_LOCAL_ENGINE_PORT)

llama-server. Reuses healthy server.

1318 (DOTTIE_LOCAL_HTTP_PORT)

dottie-local HTTP façade (desktop stack; mac-use owns :1321)

LLM_REASON=1

enable thinking on ask (default off)

LLM_MODEL / DOTTIE_LOCAL_MODEL

GGUF path or HF repo

HTTP

Method

Path

Notes

GET

/health

Engine + façade

GET

/api/tags

Ollama-shaped { models: [{ name }] } for desktop

POST

/v1/local/complete

Buffered { message } → { text }

POST

/v1/agent/chat

Dotbot harness → { text, events }

*

/v1/chat/completions

Proxied to engine (SSE ok)

GET

/v1/models

Proxied

MCP tools

Tool

Does

complete

One-shot chat, no tools

agent

Dotbot loop with dot_* tools

npm run mcp

CLI

Command

Does

ask <text>

Stream completion (local-ai-cli drop-in)

llm-server …

Manage warm engine (local-ai-cli drop-in)

dottie-local ask

Same as ask

dottie-local agent

Dotbot tools

dottie-local start|health|stop

HTTP façade / status

dottie-local stop only kills an engine this package spawned. llm-server stop also matches llama-server on the engine port (full drop-in).

Related

License

MIT

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • Ollama MCP Server

    rawveg
    A
    license
    A
    quality
    D
    maintenance
    Enables seamless integration between Ollama's local LLM models and MCP-compatible applications, supporting model management and chat interactions.
    13
    689 npm
    176
    AGPL 3.0

  • local-llm-mcp

    Qinlupfu
    A
    license
    A
    quality
    C
    maintenance
    Enables MCP clients to scan local GGUF models, estimate VRAM and suggest GPU offload layers, manage llama-server lifecycle, and proxy OpenAI-format chats with idle auto-unload.
    9
    MIT