Project Name
Dapr Agents
Project Description
Dapr Agents is an open-source framework (Linux Foundation, CNCF) for building stateful, long-running AI agents and multi-agent systems with built-in durability. It fills a critical gap in the agent ecosystem by providing durable execution, state management, and coordination capabilities needed for production workloads.
While most agent frameworks focus on reasoning and tool use, Dapr Agents focuses on reliable execution over time. It enables developers to build both single-agent workflows and multi-agent systems that can maintain state, recover from failures, wait for external events, coordinate across agents, and run for minutes, days, or months.
The project originated from the Dapr community's experience helping enterprises deploy AI systems in production. As organizations began building increasingly complex agentic applications, they needed the same reliability, scalability, and operational guarantees that Dapr provides for distributed systems. Dapr Agents brings those capabilities directly to AI agents, allowing developers to focus on business logic while knowing the any failure is automatically recovered. Dapr Agents uses the Dapr Worfklow runtime that provides durable execution for long running operations.
The project proposes to change its name to Durable Agents if accepted.
Alignment with AAIF Mission
Dapr Agents aligns with the Agentic AI Foundation's mission by providing the open, vendor-neutral infrastructure needed to run agentic systems reliably in production. As organizations move from simple chatbots to long-running, stateful, and collaborative agents, they need standardized approaches for execution, coordination, recovery, and interoperability.
Dapr Agents complements other AAIF projects by focusing on the runtime layer of agentic systems. It provides durable execution for single-agent workflows, coordination for multi-agent systems, and integration with emerging standards such as MCP and agent-to-agent communication protocols. By building on open standards and open-source technologies, Dapr Agents helps ensure that developers can build portable, interoperable agentic applications without being locked into a specific vendor, model provider, or infrastructure platform.
Through its focus on reliability, interoperability, and production readiness, Dapr Agents advances the AAIF vision of an open ecosystem for building the next generation of AI agents.
Relation to Existing AAIF Projects
Dapr Agents complements existing AAIF projects by focusing on the execution and orchestration layer of agentic systems.
-
MCP provides a standard protocol for connecting agents to tools, data, and services. Dapr Agents enhances MCP tools beyond regular usage by adding durable execution, state management, and fault recovery, making MCP-powered applications stateful, reliable, and capable of running long-lived workflows that survive failures and restarts.
-
Goose provides an open-source agent framework and developer experience for building AI agents. Dapr Agents complements this by providing infrastructure for stateful execution, fault recovery, multi-step workflows, and coordination across multiple Goose agents.
-
Agentgateway provides security, governance, observability, and policy enforcement for agent, MCP, and LLM interactions. Dapr Agents complements these capabilities by providing orchestration, durable execution, and coordination for the workloads that flow through those governed interactions.
Example Use Cases and Evidence of Adoption
Zeiss (Optics and precision engineering company 46k+ employees) - Uses Dapr Agents in production for high scale document and image analysis - see KubeCon EU 26'
RSM (Tax and auditing company, 56k+ employees) - Uses Dapr Agents in production for data querying and discovery agents
Uniphar (Largest pharma in Ireland, 3.5k+ employees) - Uses Dapr Agents to orchestrate a multi-agent exchange in charge of maintaining internal platform infrastructure
Technical Committee Sponsor (if identified)
No response
GitHub Repository URL
https://github.com/dapr/dapr-agents
License
Apache 2.0
Governance Model
Dapr Agents is currently a sub-project in the Dapr organization, under Linux Foundation and the CNCF. The project has a set of dedicated maintainers and falls under the Dapr Steering & Technical Committee.
Current STC member companies for Dapr:
- NVIDIA
- Microsoft
- Tencent
- Diagrid
- Dotmatics (Siemens)
- Innovian
CI/CD & Release Workflow
Dapr Agents is developed as an open-source Python project in GitHub under the Dapr organization. The project uses standard Python packaging with pyproject.toml, uv for dependency management, and supports modern Python versions. The repository includes source code, documentation, examples, quickstarts, schemas, and automated tests.
The project is tested through GitHub Actions. Pull requests run linting, formatting checks, schema validation, type checking, and unit tests across multiple Python versions. Integration tests are also available and can be run through GitHub Actions, install dependencies, and validate quickstart scenarios against a real environment.
Releases are tag-based. When a version tag is pushed, GitHub Actions builds the Python package and publishes it to PyPI. The project has an active release history, with 25 GitHub releases listed with the latest release v1.0.3 from May 19, 2026. Releases are made as needed based on project changes, bug fixes, and new capabilities.
Public-Facing Contribution Process for Specifications
Contributions of any kind are made available through the development and contribution guidelines.
Publicly Accessible Issue Tracker
https://github.com/dapr/dapr-agents/issues
External Project Dependencies
Maintainers & Contributors
Roberto Rodriguez @Cyb3rWard0g - NVIDIA
Samantha Coyle @sicoyle - Diagrid
Casper Nielsen @CasperGN - Diagrid
Leadership Team & Decision Process
The project follows the maintainer governance model for reviewers/maintainers as set by the Dapr project: https://github.com/dapr/community/blob/master/community-membership.md#community-membership
Final decisions are made by maintainers (listed above), conflict resolution is done through the Dapr Steering & Technical Committee.
Roadmap
Near-term (next 1–2 releases)
- Add Human In The Loop support through Workflow integration following upstream dapr design pattern.
- Add Dapr-Agents native Mistral LLM chat client
- Add Agent Executor Base allowing extension of executors from LLM Providers to agent SDKs like claude-agent-sdk
- Add claude-agent-sdk executor
- Add streaming response support for both single agent & multi-agent scenarios
Medium-term (3–6 months)
- Add agent runtime hooks to allow plugin system (this allows extending the agent runtime with plugins that can be maintained outside the dapr-agent core)
- Extend hot-reloading functionality enabling the agent to update its configuration during execution for near-zero downtime for agent maintenance (only dependency updates need restart)
- Add agent support for OAuth based invocations including exchanging to OBO tokens to pass into the Workflow Lineage to enable downstream consumers to validate who the agent works on-behalf-of
Long-term (6–12 months)
- Add support for sandbox execution to allow the agent to offload execution of untrusted activities, or file read/write, in a runtime-safe environment
- Implement Reliability & Accuracy metrics from AAIF WG
Security
The project uses GitHub Actions for automated validation, including testing and quality checks on every change. Dependencies are managed through standard Python package management tooling, and releases are produced through automated CI pipelines. Security issues can be reported through a documented security process.
The project also benefits from regular dependency updates, CVE management, security reviews, and an active maintainer community with experience operating mission-critical enterprise workloads. Dapr Agents has undergone extensive security scrutiny through its adoption across regulated industries including financial services, manufacturing and more.
Website URL
https://github.com/dapr/dapr-agents
Documented Governance Practices (if any)
https://github.com/dapr/dapr-agents/blob/main/GOVERNANCE.md
Links to Social Media Accounts
No response
Trademark and accounts
Details of Existing Financial Sponsorship
The project is already under the Linux Foundation. Microsoft sponsors CI/CD test environments.
Infrastructure Needs or Requests
No current needs.
Additional Information
While Dapr Agents is built on top of Dapr (in the same way the Envoy AI Gateway builds on top of Envoy) and benefits from the CNCF ecosystem, the Agentic AI Foundation is a more natural home for the project because its primary focus is agentic AI rather than cloud-native infrastructure. Dapr Agents exists to solve challenges specific to long-running, stateful, and collaborative AI agents, and it closely complements other AAIF projects such as MCP, Agentgateway and Goose. By joining AAIF, Dapr Agents can evolve alongside the emerging standards, protocols, and developer ecosystems while focusing on production-grade reliability and connecting the AI-native and Cloud-Native ecosystems.
To make IP transfer easier, we are prepared to change the name of the project to Durable Agents upon acceptance.
Application contact name(s) and email(s)
Yaron Schneider - yaron@diagrid.io
Contributing or sponsoring entity signatory information
Linux Foundation / CNCF as trademarks are assigned to said entities.
Project Name
Dapr Agents
Project Description
Dapr Agents is an open-source framework (Linux Foundation, CNCF) for building stateful, long-running AI agents and multi-agent systems with built-in durability. It fills a critical gap in the agent ecosystem by providing durable execution, state management, and coordination capabilities needed for production workloads.
While most agent frameworks focus on reasoning and tool use, Dapr Agents focuses on reliable execution over time. It enables developers to build both single-agent workflows and multi-agent systems that can maintain state, recover from failures, wait for external events, coordinate across agents, and run for minutes, days, or months.
The project originated from the Dapr community's experience helping enterprises deploy AI systems in production. As organizations began building increasingly complex agentic applications, they needed the same reliability, scalability, and operational guarantees that Dapr provides for distributed systems. Dapr Agents brings those capabilities directly to AI agents, allowing developers to focus on business logic while knowing the any failure is automatically recovered. Dapr Agents uses the Dapr Worfklow runtime that provides durable execution for long running operations.
The project proposes to change its name to
Durable Agentsif accepted.Alignment with AAIF Mission
Dapr Agents aligns with the Agentic AI Foundation's mission by providing the open, vendor-neutral infrastructure needed to run agentic systems reliably in production. As organizations move from simple chatbots to long-running, stateful, and collaborative agents, they need standardized approaches for execution, coordination, recovery, and interoperability.
Dapr Agents complements other AAIF projects by focusing on the runtime layer of agentic systems. It provides durable execution for single-agent workflows, coordination for multi-agent systems, and integration with emerging standards such as MCP and agent-to-agent communication protocols. By building on open standards and open-source technologies, Dapr Agents helps ensure that developers can build portable, interoperable agentic applications without being locked into a specific vendor, model provider, or infrastructure platform.
Through its focus on reliability, interoperability, and production readiness, Dapr Agents advances the AAIF vision of an open ecosystem for building the next generation of AI agents.
Relation to Existing AAIF Projects
Dapr Agents complements existing AAIF projects by focusing on the execution and orchestration layer of agentic systems.
MCP provides a standard protocol for connecting agents to tools, data, and services. Dapr Agents enhances MCP tools beyond regular usage by adding durable execution, state management, and fault recovery, making MCP-powered applications stateful, reliable, and capable of running long-lived workflows that survive failures and restarts.
Goose provides an open-source agent framework and developer experience for building AI agents. Dapr Agents complements this by providing infrastructure for stateful execution, fault recovery, multi-step workflows, and coordination across multiple Goose agents.
Agentgateway provides security, governance, observability, and policy enforcement for agent, MCP, and LLM interactions. Dapr Agents complements these capabilities by providing orchestration, durable execution, and coordination for the workloads that flow through those governed interactions.
Example Use Cases and Evidence of Adoption
Zeiss (Optics and precision engineering company 46k+ employees) - Uses Dapr Agents in production for high scale document and image analysis - see KubeCon EU 26'
RSM (Tax and auditing company, 56k+ employees) - Uses Dapr Agents in production for data querying and discovery agents
Uniphar (Largest pharma in Ireland, 3.5k+ employees) - Uses Dapr Agents to orchestrate a multi-agent exchange in charge of maintaining internal platform infrastructure
Technical Committee Sponsor (if identified)
No response
GitHub Repository URL
https://github.com/dapr/dapr-agents
License
Apache 2.0
Governance Model
Dapr Agents is currently a sub-project in the Dapr organization, under Linux Foundation and the CNCF. The project has a set of dedicated maintainers and falls under the Dapr Steering & Technical Committee.
Current STC member companies for Dapr:
CI/CD & Release Workflow
Dapr Agents is developed as an open-source Python project in GitHub under the Dapr organization. The project uses standard Python packaging with
pyproject.toml,uvfor dependency management, and supports modern Python versions. The repository includes source code, documentation, examples, quickstarts, schemas, and automated tests.The project is tested through GitHub Actions. Pull requests run linting, formatting checks, schema validation, type checking, and unit tests across multiple Python versions. Integration tests are also available and can be run through GitHub Actions, install dependencies, and validate quickstart scenarios against a real environment.
Releases are tag-based. When a version tag is pushed, GitHub Actions builds the Python package and publishes it to PyPI. The project has an active release history, with 25 GitHub releases listed with the latest release v1.0.3 from May 19, 2026. Releases are made as needed based on project changes, bug fixes, and new capabilities.
Public-Facing Contribution Process for Specifications
Contributions of any kind are made available through the development and contribution guidelines.
Publicly Accessible Issue Tracker
https://github.com/dapr/dapr-agents/issues
External Project Dependencies
Maintainers & Contributors
Roberto Rodriguez @Cyb3rWard0g - NVIDIA
Samantha Coyle @sicoyle - Diagrid
Casper Nielsen @CasperGN - Diagrid
Leadership Team & Decision Process
The project follows the maintainer governance model for reviewers/maintainers as set by the Dapr project: https://github.com/dapr/community/blob/master/community-membership.md#community-membership
Final decisions are made by maintainers (listed above), conflict resolution is done through the Dapr Steering & Technical Committee.
Roadmap
Near-term (next 1–2 releases)
Medium-term (3–6 months)
Long-term (6–12 months)
Security
The project uses GitHub Actions for automated validation, including testing and quality checks on every change. Dependencies are managed through standard Python package management tooling, and releases are produced through automated CI pipelines. Security issues can be reported through a documented security process.
The project also benefits from regular dependency updates, CVE management, security reviews, and an active maintainer community with experience operating mission-critical enterprise workloads. Dapr Agents has undergone extensive security scrutiny through its adoption across regulated industries including financial services, manufacturing and more.
Website URL
https://github.com/dapr/dapr-agents
Documented Governance Practices (if any)
https://github.com/dapr/dapr-agents/blob/main/GOVERNANCE.md
Links to Social Media Accounts
No response
Trademark and accounts
Details of Existing Financial Sponsorship
The project is already under the Linux Foundation. Microsoft sponsors CI/CD test environments.
Infrastructure Needs or Requests
No current needs.
Additional Information
While Dapr Agents is built on top of Dapr (in the same way the Envoy AI Gateway builds on top of Envoy) and benefits from the CNCF ecosystem, the Agentic AI Foundation is a more natural home for the project because its primary focus is agentic AI rather than cloud-native infrastructure. Dapr Agents exists to solve challenges specific to long-running, stateful, and collaborative AI agents, and it closely complements other AAIF projects such as MCP, Agentgateway and Goose. By joining AAIF, Dapr Agents can evolve alongside the emerging standards, protocols, and developer ecosystems while focusing on production-grade reliability and connecting the AI-native and Cloud-Native ecosystems.
To make IP transfer easier, we are prepared to change the name of the project to Durable Agents upon acceptance.
Application contact name(s) and email(s)
Yaron Schneider - yaron@diagrid.io
Contributing or sponsoring entity signatory information
Linux Foundation / CNCF as trademarks are assigned to said entities.