Skip to content
maxjo020418Public

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

64 Commits

Folders and files

Repository files navigation

Dockerina

Agentica, now with Ollama and Docker

extra docs and changelogs here

Repository layout

  • agentica-dockerina/ - scripts to build and run containers (build.sh, run.sh, stop.sh)
    • NOTE: server & client is based on the Agentica framework's basic setup
    • client/ - frontend source and build files
    • server/ - backend source and package configuration
      • agentica - somewhat modded Agentica framework (git subtree)
  • agentica-shim/ - shim utilities (used for dev/debug)
  • LLM-tests - LLM modelfiles and ChatML tests (used for dev/debug)
.
├── LLM-tests
│   └── debug_capture (not tracked)
├── agentica-dockerina
│   ├── client
│   │   ├── dist
│   │   ├── node_modules
│   │   ├── public
│   │   └── src
│   └── server
│       ├── agentica
│       ├── build
│       ├── lib
│       ├── node_modules
│       └── src
└── agentica-shim
    └── logs (not tracked)

Prerequisites

  • Docker (required for container scripts)
  • Node.js (run or build server and client)
  • pnpm (package manager)
  • Ollama (if using local models)

Docker, Ollama host can be on a different machine

regarding LLMs

  • For 3rd parties, tested on OpenAI's gpt-4o series (should work for OpenAI's v1 API scheme)
  • For Ollama, (COT - 'Chain Of Thought' models are compatible)
    • qwen3 (14b model best overall)
    • gpt-oss (smallest 20b model, still slow)
    • llama3.1

Recommended settings are:

  • $\ge$ 12b parameters (8b has some trouble)
  • $\ge$ 8k context size (4k works too but 8k is recommended)
  • lower temp (for function call hallucination prevention if needed)

system

tested on: WSL Ubuntu 24.04, Kubuntu(Ubuntu) 25.04

  • x86_64 6-core CPU (5600x)
  • 64GB RAM
  • 12GB VRAM (4070)

Quick start

  1. Build and run Dockerina:
cd agentica-dockerina
./build.sh
./run.sh
  1. Stop Dockerina:
./stop.sh

note that this will run/stop both the server and the frontend client

Optional: manual build/start

Run the server:

cd server
pnpm install
pnpm --filter @agentica/core build
pnpm --filter @agentica/rpc build
pnpm build
pnpm start

Run the client:

cd client
pnpm install
pnpm build
pnpm start

Logs

Runtime logs are captured and stored under agentica-shim/logs/(not tracked by repo, need to create one) as JSON files.

not that it only works when it's using with the shim(agentica-shim/shim.py) to capture raw requests

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages