DBOS

We're kicking off a new series highlighting the features and integrations our team has been shipping each month. The past two months were all about durable AI. We shipped Kotlin support, workflow patching & lineage tracking, dynamic workflow scheduling, configurable alerting, a new MCP server for troubleshooting, and integrations with OpenAI Agents SDK, Pydantic AI, Vercel, and more.

Durable workflow engines typically depend on adding overhead, write amplification, and complexity. At #P99CONF, Qian Li shared a lightweight library-based engine that embeds into application code and checkpoints state directly to the database. ow.ly/uBSP50XmvKr #ScyllaDB

Bridge Languages with Databases and Workflows An idea on how to make services interoperate. It’s not uncommon for a service in one language to manage tasks written in another language. For example, you might implement a document processing pipeline in Python, but enqueue documents from a TypeScript web server. The typical solution: add a middle layer like a message broker or RPC service. This works, but adds infrastructure and complexity. A radical alternative: build your services with durable workflows. Then they store their state in your database, and all of your services, in any language, can talk to your database. This means you can submit a job to another service with a simple INSERT, and check its progress with nothing more than a SELECT. But how do you do this practically? How do you build database-backed workflows in a way that interoperates across languages, given that neither languages nor databases are built for this? We wrote about this design pattern, how we implemented it in DBOS, and how you can use it in your applications. 👇

Our next South Bay Systems meetup will be on March 31. We've got two awesome deep-dive talks: - @ssougou will present deconstructing consensus and its application to Multigres - @stuhood will talk about how modern full-text search is evolving toward columnar systems (and the tricky optimization challenges) Food and beverages will be provided, courtesy of our hosts, @Snowflake Come hang out with the systems crowd 👇

Read the blog post here: dbos.dev/blog/making-la…

Applications already depend on a database (usually Postgres). So why not use the database itself as the bridge between languages? We just published a new blog post on how we tackled language interoperability without adding another layer of infrastructure. The idea is simple: A client in one language connects directly to the system database of an application written in another language, and exchanges data through workflows, messages, events, and streams. No message brokers. No RPC glue. Just Postgres as the coordination layer. Curious how it works 👇

Cool new feature: you can now enqueue workflows (and send messages to them) directly from PostgreSQL! This is really useful for starting workflows in response to database events like triggers, or for starting a workflow as part of a larger database transaction. Under the hood, this leverages the new infrastructure we built for cross-language interoperability (such as a portable JSON-based serialization format): after all, PL/pgSQL is just another programming language! Check out the docs: 👇

Just wrote a new technical deep-dive blog post! This post goes into detail about how we built dynamically configurable cron schedules. The main idea is to store schedules in the database so it can act as a “control plane” to coordinate scheduled workflows running on different workers. There are a few cool tricks that make the implementation work, including using workflow IDs as idempotency keys so multiple workers don’t do the same work, and adding jitter to prevent a “thundering herd” of database contention every time a schedule fires at scale. Check it out! 👇

I’m giving a talk today at @ScyllaDB Monster Scale Summit! Going to be talking about the challenges in building a scalable decentralized system, with a particular focus on scaling decentralized workflow execution. Check it out! 👇

Also released another much-requested feature, DBOS.waitFirst. This does something simple but powerful: it lets you durably wait for the first of several workflows to complete. It’s useful if you need to enqueue many tasks and process the results of each task as soon as it completes instead of waiting for them all to finish.

We’ve gotten a ton of requests for interoperability: allowing DBOS apps in different languages to communicate. It took a ton of work, but really happy to announce that we just released interoperability, starting in TypeScript, Python, and Java. Now, you can use a client in any of those languages to connect to an app in any other language and enqueue, manage, or send messages to workflows. For example, you can have your TypeScript web backend enqueue work in your Python or Java cluster. The key piece that makes this work is a unified JSON-based serialization layer (built on top of the already-shared system database schema) you can opt into that makes all DBOS implementations “speak the same language.” More info in the docs: 👇

@llama_index Excited for this integration!

Creating agent workflows and architecting the logic is one thing, making them durable, fail-safe, and scalable is another👇 New integration for durable agent workflows with @DBOS_Inc execution - Make sure your agents survive crashes, restarts, and errors without writing any checkpoint code. 🔄 Every step transition persists automatically - workflows resume exactly where they left off ⚡ Zero external dependencies with SQLite, or scale to multi-replica deployments with Postgres 👯‍♀️ Built for replication - each replica owns its workflows, with Postgres coordinating across instances 💤 Idle release feature frees memory for long-running workflows waiting on human input 🛡️ Built-in crash recovery detects and relaunches incomplete workflows automatically This integration with DBOS removes all the manual snapshot work from durable workflows. Just pass a DBOS runtime to your workflow and get great reliability — whether you're running a single process or multiple replicas in production. Learn how to build durable agents on our new docs: developers.llamaindex.ai/python/llamaag…

Read the full update here: dbos.dev/blog/dbos-new-…

We're kicking off a new series highlighting the features and integrations our team has been shipping each month. The past two months were all about durable AI. We shipped Kotlin support, workflow patching & lineage tracking, dynamic workflow scheduling, configurable alerting, a new MCP server for troubleshooting, and integrations with OpenAI Agents SDK, Pydantic AI, Vercel, and more.

I’m starting a new blog series: a weekly deep dive on a new feature, what problem it solves, and how it works. The idea is to put out some more technical content showing how things really work under the hood and also highlight how fast the team ships! The first entry is about the new DBOS MCP server, discussing both why MCP for workflows is useful (agentically debug your programs!) and also how you actually build an MCP server in the first place, including how to solve tricky problems like auth. Check it out! 👇

Read the blog post here: dbos.dev/blog/mcp-agent…

Coding agents are already great at writing durable workflows. But this is the easy part. Operating them, especially when things break, is much harder. If multiple workflows fail with unexpected errors: - How do you debug? - How do you find the root cause? - How do you safely test and apply a fix? That's why we built an MCP server for DBOS. Check out our latest blog post on DBOS MCP: Agent-Native Operations for Durable Workflows.

What makes building an open-source project special is seeing a community grow around it. That’s why I love our monthly user groups, to meet the community and learn what people are building with DBOS. Tomorrow, we’re going to do a deep dive into durable queues, all the advanced patterns they support, how they work, and what problems our users are solving with them. This will be more technical than our previous user groups, so hopefully you all like a twist on the format! Hope to see you there! Signup link below 👇

Cool new feature: Dynamic workflow schedules! The idea is to store workflow schedules in Postgres so you can dynamically create, view, update, and delete them at runtime, as well as (coming soon) view them from a dashboard. This lets you do things like dynamically create new schedules when you onboard customers to perform periodic work for those customers, like regular data syncs.

AI agents that survive crashes 🔧 How? Pydantic AI + DBOS. Full write up on our latest blogpost, written by @qianl_cs pydantic.dev/articles/pydan…