Agents often execute tasks of unpredictable duration.
Kodosumi leverages Ray to reliably manage these complex, long-running workflows, that can run on industry-grade infrastructure.
Optional category/Tagline
Handling Bursty Agent Traffic
Agent workloads can spike unexpectedly.
Kodosumi smoothly scales horizontally across a Ray Cluster, ensuring consistent performance even under intense load.
Optional category/Tagline
Debugging Complex Agentic Workflows
Complex agents require transparent observability.
Kodosumi provides built-in real-time insights and detailed logging via the Ray dashboard to easily diagnose and resolve issues.
Optional category/Tagline
Integrating Open Source Tools & LLMs
Teams want flexibility without lock-in.
Kodosumi offers an open, framework-agnostic platform designed specifically for seamless integration with any existing LLMs (including self-hosted) and agent frameworks.
Sokosumi Marketplace is the customer-facing storefront and agentic platform, where end-users can hire agents. Publish your agents and earn when those are used.
Kodosumi is a pre-configured runtime to build, deploy, and scale AI agents using Ray, Litestar and FastAPI.
Do I need Ray expertise?
No. Kodosumi simplifies Ray deployment. If you know how to build Agents with Python, that should be enough.
Is Kodosumi production-ready?
It depends. It’s built on Ray, Litestar and FastAPI - all trusted at enterprise scale. That said, we are still developing the framework, meaning some concepts may be a subject of change.
Can I use my existing AI models/agents/workflows/business logic?
Absolutely. Kodosumi gives you full freedom of choice in terms of underlying agentic and business logic. Use any LLMs, vector stores, AI frameworks, SDKs, then wrap them in a scalable Kodosumi runtime!
Does Kodosumi lock me into a vendor?
No vendor lock-in. Kodosumi is open-source and does not enforce developers to use any particular LLM vendors, agentic frameworks or SDKs.
Why both FastAPI and Litestar?
FastAPI powers agent endpoints; Litestar runs the admin interface and core services.
Can I configure my runtime?
Yes. Kodosumi uses a simple, minimal YAML file to configure deployable runtime alongside with config.py file which allows admin panels and development environment.
Does Kodosumi support Kubernetes or Docker?
Yes. Deployments are consistent across Kubernetes, Docker, or bare metal.
Is real-time monitoring available?
Built-in. The web panel streams real-time events and lets you replay runs.
What’s the pricing model?
It’s free and open-source. Run it on-premises, locally, or in any cloud.
# https://www.kodosumi.io/ llms-full.txt## AI Agent Deployment [home](https://www.kodosumi.io/)**New update** · Message about what’s new# Run AI Agents at Scale, Reliable and FastThe distributed runtime environment that manages and executes agentic services at enterprise scale.[Getting Started](https://docs.kodosumi.io/) [GitHub](https://github.com/masumi-network) [Discord](https://discord.com/invite/aj4QfnTS92)[](https://www.kodosumi.io/#)Framework AgnosticNewsletterSign up for updatesWelcome to the AI Agent Economy. You’ll now receive network news.Oops! Something went wrong while submitting the form.## Reasons for KodosumiCompetitive Analysis### Deploy locally, run pipelines in parallel, fine-tune - Kodosumi scales effortlessly thanks to [Ray](https://www.ray.io/).Instant Observability### See what is happening with built-in real-time monitoring via [Ray](https://www.ray.io/) dashboard.Zero Lock-In### Keep full control - Kodosumi is open source, framework agnostic, and portable across platforms.Minimal Configuration Overhead### Only a Single YAML config file required to directly deploy agents on Kodosumi.Natively Agentic### Focus on building your AI services, including long running ones, and let Kodosumi do the work behind the scenes!Learn how to build your own agent[See Details](https://www.kodosumi.io/#)## Key Challenges Solved by KodosumiOptional category/Tagline### Running Long-Lasting Agents- Agents often execute tasks of unpredictable duration. - Kodosumi leverages Ray to reliably manage these complex, long-running workflows, that can run on industry-grade infrastructure.Optional category/Tagline### Handling Bursty Agent Traffic- Agent workloads can spike unexpectedly. - Kodosumi smoothly scales horizontally across a [Ray](https://www.ray.io/) Cluster, ensuring consistent performance even under intense load.Optional category/Tagline### Debugging Complex Agentic Workflows- Complex agents require transparent observability. - Kodosumi provides built-in real-time insights and detailed logging via the Ray dashboard to easily diagnose and resolve issues.Optional category/Tagline### Integrating Open Source Tools & LLMs- Teams want flexibility without lock-in. - Kodosumi offers an open, framework-agnostic platform designed specifically for seamless integration with any existing LLMs (including self-hosted) and agent frameworks.[README.md](https://github.com/masumi-network/kodosumi)## Getting Started01**INSTALL KODOSUMI**#### Install Kodosumi#### Kodosumi is a PyPi package. Get it with pip or uv.PyPiJavaScriptCopied`pip install kodosumi`02**Create A DIRECTORY FOR YOUR AGENTIC APPS**#### Create service home#### A directory for your agentic appsJavaScriptCopied`mkdir ./home cp -r ./kodosumi/apps/hymn ./home/` **`4`**03**PREPARE CONFIG.YAML**#### Configure the environment#### In `config.yaml` you define the Python package requirements and environment variables.config.yamlconfig.yamlJavaScriptCopied`applications: - name: hymn route_prefix: /hymn import_path: hymn.app:fast_app runtime_env: pip: - crewai - crewai_tools env_vars: OTEL_SDK_DISABLED: "true" OPENAI_API_KEY: add your key here ` **`5`**04**Start Ray**#### Start Ray as a daemon#### Change to `./home` and start Ray inside this directory so Ray can import from this directory.JavaScriptCopied`cd home ray start --head`05**Configure environment variables**#### Deploy your agents#### Deploy example services, or your own flows right away.filename.shJavaScriptCopied`cp .env.` `example` ` .env nano .env`06**Deploy example apps with Ray**#### Launch the service in your localhost Ray cluster.#### Ensure you start serve in the same directory as Ray ( `./home`).JavaScriptCopied`serve deploy ./hymn/config.yaml`07**Start Kodosumi**#### Access your service#### Finally start the kodosumi components and register ray endpoints available at [http://localhost:8001/-/routes](http://localhost:8001/-/routes).JavaScriptCopied`koco start --register http://localhost:8001/-/routes`[README.md](https://github.com/masumi-network/kodosumi)## Core ConceptsWe use a combination of terms Agents - Flows - Agentic Services to describe the system.#### Agent#### An autonomous object within the Kodosumi framework that can perform specific tasks or services.Agents can interact with other agents, process data, and execute complex workflows.#### Flow#### An automated process, workflow, or system of interconnected tasks working towards a common objective.We use the term Flow to emphasize its process-oriented nature.#### Agentic Service#### A self-contained, deployable unit within the Kodosumi framework.The Agentic Service integrates one or more Flows with required resources and configurations to deliver complete functionality.[Read Docs](https://docs.kodosumi.io/)Learn how to build your own agent[See Details](https://www.kodosumi.io/#)Free & Open-Source## Contribute, Create an Issue, DiscussKodosomi Repo is a great option for getting answers to Masumi technical questions lorem ipsum.[Visit masumi repo](https://github.com/masumi-network)Sokosumi Marketplace is the customer-facing storefront and agentic platform, where end-users can hire agents. Publish your agents and earn when those are used.[Visit Sokosumi Marketplace](https://www.sokosumi.com/)Masumi Network makes deploying your agents easy and lets you participate in Sokosumi Marketplace.[Visit masumi network](https://www.masumi.network/)Joining Discord will get you quick answers and is the best place to discuss upcoming features and products with developers and early adopters.[Visit Discord](https://discord.com/invite/aj4QfnTS92)For Developers## Monetize Your AgentsDeploy your agent on Masumi Network to participate in Sokosumi Marketplace[Masumi Network](https://www.masumi.network/) [Sokosumi Marketplace](https://www.sokosumi.com/) [Masumi Repo](https://github.com/masumi-network)## FAQ### What exactly is Kodosumi?Kodosumi is a pre-configured runtime to build, deploy, and scale AI agents using Ray, Litestar and FastAPI.### Do I need Ray expertise?No. Kodosumi simplifies Ray deployment. If you know how to build Agents with Python, that should be enough.### Is Kodosumi production-ready?It depends. It’s built on Ray, Litestar and FastAPI - all trusted at enterprise scale. That said, we are still developing the framework, meaning some concepts may be a subject of change.### Can I use my existing AI models/agents/workflows/business logic?Absolutely. Kodosumi gives you full freedom of choice in terms of underlying agentic and business logic. Use any LLMs, vector stores, AI frameworks, SDKs, then wrap them in a scalable Kodosumi runtime!### Does Kodosumi lock me into a vendor?No vendor lock-in. Kodosumi is open-source and does not enforce developers to use any particular LLM vendors, agentic frameworks or SDKs.### Why both FastAPI and Litestar?FastAPI powers agent endpoints; Litestar runs the admin interface and core services.### Can I configure my runtime?Yes. Kodosumi uses a simple, minimal YAML file to configure deployable runtime alongside with config.py file which allows admin panels and development environment.### Does Kodosumi support Kubernetes or Docker?Yes. Deployments are consistent across Kubernetes, Docker, or bare metal.### Is real-time monitoring available?Built-in. The web panel streams real-time events and lets you replay runs.### What’s the pricing model?It’s free and open-source. Run it on-premises, locally, or in any cloud.we are online## Join the Community[Discord](https://discord.com/invite/aj4QfnTS92) [Twitter](https://x.com/MasumiNetwork).avif)
.avif)