Dosu: Agentic AI for Software Projects

Dosu is an agentic AI system that transforms codebases into living knowledge bases for software teams, enabling them to ship faster and collaborate better. DBOS durable execution and durable queuing makes the system reliable and observable for the 50,000 software projects that use Dosu.

Customer
Dosu
Industry
Software, AI

About Dosu

Dosu is an AI agent that transforms your codebase into a living knowledge base, so you can focus on building instead of answering questions and maintaining docs. Dosu automatically generates and maintains documentation for your software by analyzing not just code, but also the context found in conversations, issues, tickets, and reviews. Dosu AI automation makes it possible for teams to ship faster, onboard people more quickly, collaborate better and reduce maintenance overhead. 

Today, Dosu is used by over 50,000 software projects, including rapidly growing open-source standouts like BetterAuth, LlamaIndex, Apache Airflow, and Zod.

Use Case: RAG and Agentic AI Pipeline Reliability

At the heart of the Dosu platform is a sophisticated indexing and search engine that ingests and processes information about users' projects from code, GitHub issues, Slack, documentation, and tickets. Workflows are triggered by changes to content, and also run on a scheduled basis. These workflows transform software project information into the intelligent knowledge base that makes Dosu so powerful. Dosu is mission critical to software teams, therefore the workflows have to be performant and reliable for software projects of any size.

Challenges: Workflow Complexity & Observability

Initially, Dosu built their pipelines using Celery, a popular Python task queue, and three challenges soon became impossible to ignore:

  1. Orchestrating complex workflows. Dosu's workflows load, transform, store, and index hundreds of thousands of documents per day.. When onboarding a new project, the system might need to ingest their entire codebase history, a process that can span several hours and involve hundreds of thousands of interconnected tasks. Handling GitHub rate limiting and the possibility of network disruptions further complicated development. Celery has limited support for orchestrating workflows, which made developing and troubleshooting the Dosu pipelines more difficult and time consuming.
“Celery is great at putting a task on a broker, but its orchestration primitives were really limiting especially for highly parallelized and long-running workflows. Whenever you want to do something multi-step, it starts to get ugly." - Devin Stein, Founder and CEO, Dosu.dev.

  1. Workflow and queue observability. Observability of RAG pipelines is critically important because it helps the team improve the quality of prompts and generated responses over time. Celery only monitors individual tasks, not entire workflows. When something failed, the team couldn't easily see which workflow the task belonged to or track the status of nested steps, which made troubleshooting a challenge. 

  1. Infrastructure and operations. While Dosu's main infrastructure runs on autoscaling Google Cloud Run containers, Celery requires its own dedicated cluster on Google Compute Engine with a dedicated Redis broker. As the platform scales, this architectural split creates an operational burden with two different deployment models, two sets of scaling concerns, and increased complexity across the entire stack.

Solution: Migrating from Celery to DBOS

Recognizing these challenges, Dosu's team evaluated workflow orchestration solutions before deciding on DBOS. The decision came down to simplicity. While other solutions such as Temporal required hosting and managing separate orchestrators alongside their code, DBOS offered a lightweight library that could run anywhere.

Within just a few weeks, Dosu successfully migrated their entire ingestion pipeline to DBOS and scaled it to processing 20,000 workflows per hour. DBOS was the perfect fit for their challenges because it provides:

Effortless workflow orchestration: DBOS allowed Dosu to implement their entire document ingestion pipeline as a single workflow. The system now automatically handles tens of thousands of documents concurrently and recovers from failures without losing progress.

Built-in observability: Using DBOS Conductor, the Dosu team gained real-time visibility into workflow and queue health. When tasks fail, engineers can now view errors within the complete context of the workflow, dramatically reducing debugging time.

Infrastructure consolidation: DBOS applications don't require separate orchestration servers. After migration, Dosu deployed their pipeline to the same autoscaling Cloud Run containers as the rest of their infrastructure, and stored workflow state information in their Postgres database, which eliminated architectural complexity and reduced operational overhead.

“DBOS is really simple to manage. We are a Postgres shop already. so we didn't have to spin up any new infrastructure and it also worked really well with Cloud Run. DBOS gives us durable orchestration and observability without a lot of coding and without having to run additional infrastructure.”

Results

The migration to DBOS significantly improved Dosu's operations. With built-in fault tolerance and observability, the engineering team spends less time debugging production issues and more time building features. Today, Dosu's knowledge base pipeline reliably serves over 50,000 customers, automatically running tens of thousands DBOS workflows per hour to process millions of documents.

You can see the results in action on the Dosu’s Public Space for DBOS.

Reliable backends,
built effortlessly

Discover why brands are turning to DBOS for reliable and observable programs.
Add a few annotations to your program to make it resilient to any failure.

  • Deploy in minutes, not months
  • Get built-in observability for every step
  • Recover from failure without writing extra code