kodosumi 0.8.0

kodosumi framework to execute and orchestrate agentic services safe and at scale

>kodosumi

[!NOTE]

This is an early development version of kodosumi. The documentation is under development, too. See the key concepts.

kodosumi is a runtime environment to manage and execute agentic services at scale. The system is based on ray - a distributed computing framework - and a combination of litestar and fastapi to deliver men/machine interaction.

kodosumi is one component of a larger ecosystem with masumi and sokosumi.

introduction

kodosumi consists of three main building blocks:

1. The Ray cluster to execute agentic services at scale.
2. The kodosumi web interface and API services.
3. Agentic Services delivered through kodosumi and executed through Ray.

installation

This installation has been tested with versions ray==2.44.1 and python==3.12.6.

STEP 1 - clone and install kodosumi.

git clone https://github.com/masumi-network/kodosumi.git
cd kodosumi
git checkout feature/candidate
python -m venv .venv
source .venv/bin/activate
pip install -e .

STEP 2 - start ray as a daemon.

ray start --head

Check ray status with ray status and visit ray dashboard at http://localhost:8265. For more information about ray visit ray's documentation.

STEP 3 - prepare environment

To use openai API you need to create a local file .env to define the following API keys:

OPENAI_API_KEY=...
EXA_API_KEY=...
SERPER_API_KEY=...

STEP 4 - deploy example apps with ray serve

Deploy the example services available in folder ./apps. Use file apps/config.yaml.

serve deploy apps/config.yaml

Please be patient if the Ray serve applications take a while to setup, install and deploy. Follow the deployment process with the Ray dashboard at http://localhost:8265/#/serve. On my laptop initial deployment takes three to four minutes.

STEP 5 - start kodosumi

Finally start the kodosumi components and register ray endpoints available at http://localhost:8001/-/routes.

koco start --register http://localhost:8001/-/routes

STEP 6 - Look around

Visit kodosumi admin panel at http://localhost:3370. The default user is defined in config.py and reads name=admin and password=admin. If one or more Ray serve applications are not yet available when kodosumi starts, you need to refresh the list of registered flows. Visit Routes Screen at (http://localhost:3370/admin/routes in the Admin Panel at http://localhost:3370/admin/flow. See also the OpenAPI documents with Swagger http://localhost:3370/schema/swagger.

If all went well, then you see a couple of test services. Be aware you need some OpenAPI, Exa and Serper API keys if you want to use all Agentic Services.

Stop the kodosumi services and spooler by hitting CNTRL+C in the corresponding terminal. Stop Ray serve with serve shutdown --yes. Stop the ray daemon with command ray stop.

development notes

The development notes provide an overview for various flavours on how to run and deploy agentic services.

Follow the examples:

• Function Blueprint
• Search for Armstrong Numbers and with nested remote calls
• Crew of Agents to craft a Hymn using OpenAI
• Crew of Agents to craft a Marketing Campaign using OpenAI
• Crew of Agents to craft a Job Posting using OpenAI