Basecamp — Portable AI Agent with Auto-Discovery

The Problem

You've got local AI services running — Ollama, maybe TabbyAPI or vLLM, maybe SearXNG. They work. But to actually use them together, you're either:

Opening 5 browser tabs and copy-pasting between them
Hardcoding API URLs and keys into config files
Writing glue code every time you add a service

Nothing ties your stack together. Until now.

What Basecamp Does Auto-Discovery

Basecamp is a Docker image containing Hermes Agent + Ollama. When you run it on a Docker network, it:

Scans for AI services using DNS probes, port scanning, and Docker network inspection
Identifies service types by fingerprinting HTTP responses (Ollama, TabbyAPI, vLLM, SearXNG, MCPO, and more)
Shows a connect screen where you pick which services to use
Prompts for API keys when a service needs auth — no hardcoded secrets
Generates a tavern CLI dynamically from what it found
Drops you into Hermes Agent connected to your entire stack

The Connect Screen

First run, basecamp scans your network and shows you what it found:

Discovered 5 service(s): [ 0] LLM Ollama (GGUF inference) [1 model(s)] -> http://ollama:11434 [ 1] EXL TabbyAPI / OpenAI-compatible (EXL3/EXL2) [1 model(s)] [needs auth] -> http://tabbyapi:5000 [ 2] SRC SearXNG (private search) -> http://searxng:8080 [ 3] MCP MCPO (MCP OpenAPI proxy) -> http://mcpo:8000 [ 4] UI Open WebUI (LLM workspace) -> http://open-webui:8080 PRIMARY inference engine: 1. Ollama (GGUF inference) (http://ollama:11434) 2. TabbyAPI / OpenAI-compatible (http://tabbyapi:5000) 0. Use basecamp's own Ollama Choice [1]: 2 TabbyAPI requires authentication (bearer). Enter API key (or press Enter to skip): your-key-here Save and continue? [Y/n]: Y

Subsequent runs skip this — config is saved to a persistent volume. Run tavern rediscover to re-scan.

Quick Start

# Pull the image (Docker Hub)
docker pull jpanasuk/basecamp:1.0.2

# Run against your AI network
docker run -it --rm \
  --network your-ai-network \
  --gpus all \
  -v basecamp-ollama:/root/.ollama \
  -v basecamp-hermes:/root/.hermes \
  jpanasuk/basecamp:1.0.2

# First run: connect screen appears
# Pick services, enter API keys if needed
# Then you're in Hermes Agent with tavern CLI

tavern status        # show connected services
tavern chat "hello"  # chat with EXL3 model (streaming)
tavern search "AI news" # search via SearXNG
tavern mcp           # list MCP tools
tavern models         # list available models

What It Discovers

LLMOllama
GGUF inference
Auto-discovered, no auth

EXLTabbyAPI
EXL3/EXL2 inferenceneeds API key
Prompted during connect

LLMvLLM
High-throughput inference
Auto-discovered, no auth

PXYLiteLLM
Multi-provider proxyneeds API key
Prompted during connect

LLMLocalAI
OpenAI replacement
Auto-discovered, no auth

LLMllama.cpp
Lightweight server
Auto-discovered, no auth

SRCSearXNG
Private metasearch
Auto-discovered, no auth

MCPMCPO
MCP OpenAPI proxyneeds API key
Prompted during connect

UIOpen WebUI
LLM workspace
Auto-discovered, no auth

CHTSillyTavern
Character chatneeds basic auth
Tries default, prompts if fails

Why It's Different

Feature	Other tools	Basecamp
Service discovery	Manual config	Auto-discover
API keys	Hardcoded in config	Prompted at connect
Works with any stack	Tied to one app	Stack-agnostic
Agent included	BYO agent	Hermes Agent built-in
Streaming chat	Usually no	Yes, by default
Re-scan after changes	Edit config again	tavern rediscover
Secrets in image	Often yes	Zero hardcoded keys

Architecture

┌─────────────────────────────────────────────┐ │ basecamp container │ │ ┌───────────┐ ┌──────────────────────┐ │ │ │ Ollama │ │ Hermes Agent │ │ │ │ (GGUF) │ │ (CLI AI assistant) │ │ │ └───────────┘ └──────────────────────┘ │ │ ┌──────────────────────────────────────┐ │ │ │ discover.py │ │ │ │ - scans network for AI services │ │ │ │ - connect screen TUI │ │ │ │ - generates tavern CLI dynamically │ │ │ └──────────────────────────────────────┘ │ │ ┌──────────────────────────────────────┐ │ │ │ tavern (auto-generated) │ │ │ │ - chat, search, mcp, models, status │ │ │ │ - streaming responses │ │ │ │ - rediscover on demand │ │ │ └──────────────────────────────────────┘ │ └─────────────────────────────────────────────┘ │ │ │ ▼ ▼ ▼ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ TabbyAPI │ │ Ollama │ │ SearXNG │ │ (EXL3) │ │ (GGUF) │ │ (search) │ └──────────┘ └──────────┘ └──────────┘ │ │ ▼ ▼ ┌──────────┐ ┌──────────┐ │ MCPO │ │Open WebUI│ │ (tools) │ │ (UI) │ └──────────┘ └──────────┘

Related Projects

Tabby Tavern Stack — the full 6-service local AI infrastructure (TabbyAPI, SillyTavern, Open WebUI, Ollama, SearXNG, MCPO) → huggingface.co/jpanasuk/tabby-tavern-stack
Basecamp — this project, the portable agent container → huggingface.co/jpanasuk/basecamp