Open Executive: An Open-Source AI Virtual Leadership Team
This analysis explores the architecture and operational considerations for engineers deploying an open-source AI system designed to act as a company's executive team.
AI-Powered Executive Orchestration
SenteLabsAI has released Open Executive, an open-source AI system that functions as a virtual executive team. According to github.com, this system aims to provide senior advisory capabilities with expertise equivalent to an MBA, tailored to a specific business. It achieves this by presenting a single, unified executive voice to the user, even though its internal operations are handled by eight distinct specialist AI agents. These agents cover key business functions such as strategy, finance, human resources, legal, operations, marketing, product, and board communications.
The system’s architecture relies on an Executive Orchestrator, powered by Anthropic’s Claude-Sonnet-4-6, to process user messages. This orchestrator then directs requests to the relevant specialist agents in parallel. Each specialist agent retrieves context from two layers: built-in MBA-level knowledge and company-specific documents uploaded by the user, both stored in ChromaDB. This structured approach helps ensure that responses are both knowledgeable and relevant to the company’s unique situation.
Engineering the System for Coherence and Persistence
For engineering teams considering such an application, the Open Executive project demonstrates several crucial architectural choices for maintaining a coherent and persistent AI experience. The system employs episodic memory, recording key decisions and advice in SQLite after each interaction. This ensures that the executive persona ‘remembers’ past recommendations across sessions, a vital feature for ongoing business guidance.
Prompt caching is also implemented, with the executive persona, company profile, and knowledge index cached separately to achieve high hit rates. This is an important detail for optimising performance and reducing costs associated with large language model calls. The system also includes a scheduler to proactively manage follow-ups and time-sensitive actions, although the repository notes that the API must run as a single instance to prevent scheduled actions from double-firing.
The project’s tech stack is built on Python 3.11 with FastAPI for the backend, Next.js 15 for the web UI, and utilises Anthropic’s Claude API for its LLM backbone. It leverages ChromaDB for vector storage and SQLite for episodic memory, all packaged with uv as the package manager. This combination offers a robust foundation for building and deploying a complex AI application.
Flexible Deployment and Integration for Teams
Open Executive offers significant flexibility for deployment and integration, which is a key consideration for technical teams. The project supports deployment to Fly.io, with separate development and quality assurance environments managed through GitHub Actions. The repository specifies a single-instance deployment for the API due to the scheduler’s design, a critical operational detail for engineers to note.
Crucially, the system is designed to run against OpenAI-compatible local servers like Ollama or LM Studio, either exclusively or in a hybrid setup alongside Anthropic’s API. This capability allows teams to manage costs, address data privacy concerns by keeping sensitive data on-premises, and experiment with different models. According to github.com, this local model support applies to all specialists, allowing for granular control over which models handle specific executive functions. The system also integrates with various communication platforms, including Web UI, Slack, Email, Telegram, Google Chat, Discord, and a command-line interface, providing multiple access points for users.
Developers looking to extend the system can add new specialist agents by following a defined process that includes creating a new agent file, adding a system prompt, registering it with the orchestrator, and providing evaluation scenarios. This structured contribution model, combined with an Apache 2.0 license, makes Open Executive a practical foundation for bespoke virtual executive applications.