JUNE 5 2025
The path to Modus v1.0: the agent runtime for production
Modus v1.0 introduces the first production-ready agent runtime with actor-based architecture, streaming capabilities, MCP integration, and graph memory. Built on WebAssembly for sub-10ms cold starts and automatic scaling


At Hypermode, we’re building Modus to be the backbone of agentic systems, where intelligent, autonomous agents operate as first-class citizens in your stack. Since launch, Modus has enabled teams to deploy, orchestrate, and scale AI agents in hours, not weeks. As we approach v1, we’re doubling down on the technical foundations that set Modus apart, and we’re excited to share what’s coming next.
While Modus is used in demanding production environments today, the v1 milestone represents a completion of the first chapter of our vision for agent orchestration.
Why an agent runtime, not just a library?
Most frameworks in the agent ecosystem are libraries–bundles of code you import and run inside your own process. Modus is fundamentally different: it’s a runtime, purpose-built for agents. This distinction matters deeply for teams building at scale.
As we chart the course to v1, it’s evident that a runtime-based approach is the only viable path to effectively coordinate across models, tools, and data in production. Each agent instance gets a dedicated execution environment, sandboxed for resiliency, scalability, and security. When building with Modus, you can be confident that your most popular agents can scale to meet future demand.
Furthermore, Modus leverages WebAssembly to provide the dynamic compute needed for sporadic and increasingly asynchronous agent interactions that may span hours or days. Agents automatically release resources when not in use and resume from where they left off with less than 10ms of overhead.
And with the runtime the core of the open source Modus project, you have ultimate control for where to host your agents.
What’s coming by Modus v1
We’re approaching v1 as a milestone, not the destination. The groundwork for these features has been laid since the early iterations of Modus and we’ll continue to release improvements incrementally in the v0.x series for continued accessibility by the community. Rather than a feature dump, the release of v1 will represent foundational stability for the next phase of Modus’ development.
Agents as actors
The path to Modus v1 starts with the adoption of an actor framework for agents. Instead of treating agents as simple function calls or background tasks, the actor framework implements each instance of your agent as an independent actor with its own mailbox, lifecycle, and execution context. This architecture enables true concurrency and fault isolation, as well as paving the path for agent-to-agent communication. If an agent instance fails, it automatically restarts, ensuring resilience and reliability at scale. This actor-based approach draws inspiration from proven systems like Erlang, but is tailored specifically for the needs of modern AI agents.
Tasks as learned skills
Tasks codify the skills an agent has learned, turning human-led training into one-shot instructions that can be executed on demand. Each task represents a concrete capability that is modular, composable, and reusable. With future support for schedules, webhooks, and email triggers, tasks enable agents to act asynchronously and autonomously. This approach also underpins Agent2Agent protocol support, allowing agents to delegate and share skills seamlessly across workflows.
Stream in, stream out
Another major advancement is the introduction of streaming capabilities, both for inputs and outputs. Agents will be able to process data as it arrives, whether from real-time sources like Kafka, WebSockets, or HTTP streams. Outputs can be streamed as well, enabling agents to emit results, logs, or intermediate states continuously. This streaming model is essential for building responsive conversational agents, real-time data pipelines, and any application where low-latency, continuous interaction is required.
Model Context Protocol client and custom tools
We’re also evolving Modus into a first-class consumer of Model Context Protocol (MCP) servers. With these integrations, Modus agents will be able to dynamically discover and invoke tools registered on any MCP-compliant server, unlocking access to a growing ecosystem of external capabilities—from proprietary APIs to open-source utilities. Authentication and credential management will continue to be handled seamlessly by the runtime, so agents can securely interact with third-party services without risk of credential leakage. Beyond consuming existing tools, Modus will make it straightforward to define and register custom tools, allowing teams to extend the agent’s reach with domain-specific actions, data sources, or workflows.
Long-term graph memory
Memory is a critical part of any intelligent agent that improves over interaction. Rather than relying on flat key-value stores or simple document databases, Modus agents will have access to a graph model that captures dynamic relationships between entities, events, and knowledge. This enables agents to perform context-aware reasoning, retrieve relevant information efficiently, and build up long-term memory over time. The underlying storage engine in ModusGraph builds on Dgraph, the most popular open source graph engine.
ThreadKit
Finally, we’re bridging the gap between backend intelligence and user experience with ThreadKit, a toolkit of UI components for embedding agents directly into apps. ThreadKit provides pre-built, customizable components and APIs for multi-user agent interaction, all designed to connect seamlessly with Modus agent streams. Developers can easily integrate these components into popular frontend frameworks like React, enabling real-time, interactive experiences powered by Modus agents. The APIs and components are extensible and themeable, so teams can tailor the look and feel to match their brand and workflow.
Ready to build?
Each of these features is being developed with a focus on reliability, scalability, and developer experience. Together, they represent a major leap forward for building agents, and set the stage for Modus to be the robust, flexible, and production-ready runtime required for the next generation of agents.
We’re rolling out these features incrementally, starting with agents as actors in v0.18.
As we approach Modus v1, we’re looking for developers and teams eager to experiment, prototype, and help define the future of agent runtimes. If you’re interested in collaborating directly with the Hypermode engineering team, testing new features, or building advanced agent interactions, join the conversation on Discord.
Need help with your use case? See how a Hypermode Agent Lab could bring your concepts to life.