Back to Blog

Soorma Core v0.7: The 'No-Magic' Guide to Event-Driven Agents

January 23, 2026•Soorma Engineering

We're building Soorma in public, and we want your feedback early.

The short version: Soorma isn't shippable yet (v1 is coming later in 2026), but it's absolutely learnable and prototypable right now. We're actively developing v0.6 (Foundations) and v0.7 (Memory) with versioning to track changes as we iterate.

Why share now? We want to know: Is this helpful? Do the learning paths make sense? Does the DisCo framework and Soorma DevEx resonate with how you think about building agents? Your feedback now helps us refine the platform so when v1 launches, you can hit the ground running.

Here's what's available for local learning and prototyping:

  • v0.6.0 (Foundations): Event-driven choreography patterns—see how agents collaborate through events.
  • v0.7.0 (Memory): CoALA memory framework—working, semantic, and episodic memory in action.

We've created a Learning Path with progressively complex examples. Here's what you can explore on your local machine.

Getting Started:

# Clone the repository
git clone https://github.com/soorma-ai/soorma-core.git
cd soorma-core

# Install from source (recommended during pre-launch)
pip install -e sdk/python

# Start infrastructure
soorma dev --build

Why install from source? During pre-launch with breaking changes, installing from source keeps your SDK and infrastructure synchronized when you pull updates.

Part 1: The Foundations (v0.6.0)

Most agent frameworks accidentally build a "monolith in disguise" by forcing a central Orchestrator to micromanage every step. Soorma is different. We use Choreography: agents react to events, not commands.

1. The "Hello World" That Actually Scales

View Code: 01-hello-world This isn't just a print statement. In this example, you'll learn the full lifecycle: booting a Worker, connecting to the Event Bus, and handling your first action-request. It’s the minimal viable agent.

2. Decoupling with Simple Events

View Code: 02-events-simple We move beyond simple Request/Response. Here, agents publish business-facts (like "File Uploaded") without knowing who is listening. This is how you build systems where you can add new agents without breaking old ones.

3. Structured Reasoning (No More String Parsing)

View Code: 03-events-structured Events shouldn't be guessed; they should be typed. This example demonstrates how to use Pydantic models to define event schemas. The result? Your LLM knows exactly what arguments to generate, and the runtime validates them before they ever hit the wire.

Part 2: The Memory Systems (v0.7.0)

An agent without memory is just a fancy function call. We've implemented the CoALA framework (Cognitive Architectures for Language Agents) to give your agents three distinct types of recall in local development.

4. Working Memory: The "Shared Brain"

View Code: 04-memory-working How do two agents collaborate on a task without passing a massive context window back and forth? Working Memory. This example shows how to maintain a shared, plan-scoped state that allows agents to "pass the baton" seamlessly.

5. Semantic Memory: RAG Without the Mess

View Code: 05-memory-semantic This is your long-term knowledge store. We show you how to implement a RAG (Retrieval-Augmented Generation) pattern where agents autonomously query the database to ground their answers in facts, using pgvector under the hood.

6. Episodic Memory: The Holy Grail

View Code: 06-memory-episodic This is what makes a "Digital Employee" different from a chatbot. This example combines all three memory tiers to create an agent that remembers you. It recalls past interactions, user preferences, and previous decisions to build relationships over time.

What's Next?

We're actively working on Stage 3: Advanced Agent Models with complex Tool use and dynamic worker patterns. Expect breaking changes as we build toward v1.

Your feedback matters: Try the learning paths, prototype your agent ideas, and tell us what works (or doesn't). Did the examples help you understand event-driven agents? Does the DisCo framework make sense for your use cases?

Clone the Repo & Start Learning →

When v1 ships, you'll be ready to build.