Declarative Agents: Profile-Driven LLM Agent Runtime in Go
This repository profile is provided by osrepos.com, an open source repository discovery platform.

Summary
Declarative Agents by Nokia Bell Labs offers a profile-driven runtime and design patterns for building tool-augmented LLM agents. It allows defining agents, their tools, states, and transitions via YAML profiles, eliminating the need for code changes for workflow alterations. This Go-based framework promotes flexible and dependable agent development.
Repository Information
Topics
Click on any tag to explore related repositories
Use at your own risk
OSRepos shares public repositories for knowledge and discovery only. Any installation, execution, configuration, or use of code from these repositories is the user's own responsibility. Always review the repository, source code, dependencies, licenses, and security implications before running or installing anything. OSRepos is not responsible for issues, damages, or losses resulting from third-party repositories.
Introduction
Declarative Agents is a project from Nokia Bell Labs providing a profile-driven runtime and design patterns for declarative, tool-augmented LLM agents. At its core, an agent is defined by a YAML profile, which is then interpreted and executed by a single Go runtime. This approach aims to simplify the development and modification of complex agent workflows.
Why Use It and Key Benefits
Traditional agent frameworks often couple workflow logic tightly with code, meaning any change to an agent's loop or behavior necessitates a binary rebuild. Declarative Agents addresses this by externalizing the agent's definition into a YAML profile. This profile specifies the agent's tools, states, transitions, signals, and budgets.
The primary benefit is flexibility: altering an agent's workflow becomes a simple configuration edit, and creating a new agent does not require compiling new binaries. This significantly speeds up iteration and deployment. The project also includes a companion white paper, Design Patterns for Declarative Agents, which outlines eleven patterns for building reliable agents using this declarative methodology. The modular structure, with components like agent-core for the runtime engine and applications/catalog for reusable blocks, further enhances its utility.
Installation
To get started quickly, you'll need Go and Mage installed. Navigate to one of the packaged applications and run it:
cd applications/agent-architecture
mage run
This command starts the canonical documentation-curator and serves its UI.
Examples
You can validate and run an agent directly from its YAML profile. First, build the core binaries and set the catalog root:
cd agent-core
mage build # compiles cmd/ binaries into bin/
export AGENT_CATALOG_ROOT=../applications/catalog
bin/agent --profile "$AGENT_CATALOG_ROOT/agents/executor/profile.yaml" --validate-config
bin/agent --profile "$AGENT_CATALOG_ROOT/agents/executor/profile.yaml" --core-root .
The --validate-config flag reads the profile, machine, and tool definitions, then exits, performing a quick boot smoke test. You can swap executor for any other profile found under applications/catalog/agents/.
Links
- GitHub Repository: Nokia-Bell-Labs/declarative-agents
- Contact: Petar Djukic
Related repositories
Similar repositories that may be relevant next.

DeclarAgent: Declarative Runbook Executor for Safe AI Agent Workflows
September 12, 2026
DeclarAgent is an innovative declarative runbook executor specifically designed for AI agents. It enables agents to validate, dry-run, and safely execute multi-step YAML workflows. This tool provides a structured, auditable, and secure way for LLM agents to interact with real CLI workflows, enhancing their operational safety and reliability.

tunnel-client: Securely Connect Private MCP Servers to OpenAI Services
September 2, 2026
The `tunnel-client` is an OpenAI-developed client that enables secure connections between private or localhost Model Context Protocol (MCP) servers and OpenAI services like ChatGPT, Codex, and AgentKit. It allows developers to integrate their internal MCP servers without exposing them to the public internet, ensuring enhanced security and operational control. This tool is ideal for scenarios requiring secure access from private networks to OpenAI's powerful AI models.
deja-vu: Retroactive Memory for AI Coding Agents
September 2, 2026
deja-vu is a powerful local-first tool that provides retroactive memory for AI coding agents, indexing past coding sessions from various agents, even those from before installation. This Go binary allows agents to recall relevant information without needing an LLM or embeddings by default. It enhances agent performance by providing context at the point of action, preventing repeated mistakes and improving efficiency.

Agent-Sandbox: Enterprise-Grade Sandbox for AI Agents on Kubernetes
August 28, 2026
Agent-Sandbox provides an easy-to-use, enterprise-grade sandbox platform for AI Agents. It allows agents to securely run untrusted LLM-generated code, perform browser and computer use, and deploy websites with multi-session and multi-tenant isolation. This self-hosted solution wraps a robust Kubernetes foundation behind a simple RESTful API, making it accessible for developers.
Source repository
Open the original repository on GitHub.