OpenExecutive open-sources a self-hostable AI executive team

Sente Labs open-sourced OpenExecutive on 26 August 2026, a self-hostable AI system that acts as a company's executive team. The release doubled as a Hacker News launch: a thread titled "CEO fired developers to make room for AI. Developers create open source AI CEO" reached 788 points and 538 comments. Within about a day the repository had 1.2k stars, 71 forks and 16 commits.

OpenExecutive is an open-source, self-hostable AI system from Sente Labs that stands in for a company's executive team. Eight specialist agents cover strategy, finance, HR, legal, operations, marketing, product and board communications, but a single Executive Orchestrator handles every reply, so the user only ever talks to one consistent executive voice.

What it does

• Eight roles, one voice. Chief Strategy Officer, Chief Financial Officer, Chief HR/People Officer, General Counsel, Chief Operating Officer, Chief Marketing Officer, Chief Product Officer and a Board Communications Director all report through a single Executive Orchestrator. The user only sees that orchestrator's replies, never the specialists behind it.
• Two model tiers. The orchestrator and most specialists run claude-sonnet-4-6, routing to specialists through parallel tool calls. The CSO, CFO, General Counsel and Board Communications Director switch to claude-opus-4-7 with extended thinking for harder calls.
• Retrieval and memory. Each specialist pulls from built-in MBA-level knowledge stored in ChromaDB, plus a separate collection of the company's own uploaded documents. After every reply, a background claude-haiku-4-5 pass logs decisions and advice to SQLite, and the next session opens with a summary of what it recommended before.
• Caching and scheduling. Persona, company profile and knowledge index cache separately; Sente Labs says this reaches an 85% cache hit rate after the first few turns. A built-in scheduler claims due actions with an UPDATE ... RETURNING query, which is why the API has to run as a single instance instead of scaling horizontally.
• A security layer underneath. OpenExecutive already runs on extensible-mcp, a separate Sente Labs project that checks every server load, tool search and call in code the model can't reach or rewrite.
• Open and self-hostable. Python 3.11, FastAPI, ChromaDB, SQLite and a Next.js 15 UI, released under Apache 2.0 with Fly.io configs for the API, UI and QA environments, plus an in-process Discord bot with /ask and /today slash commands.

Why it matters for agent builders

For anyone building multi-agent systems, OpenExecutive doubles as a public case study, the kind of product AI Agents Listing tracks alongside coding agents and MCP servers.

• Cost-tiered reasoning. Routine specialists run the cheaper claude-sonnet-4-6. The four roles most likely to produce an expensive mistake, CSO, CFO, General Counsel and Board Communications, get claude-opus-4-7 with extended thinking instead: a concrete cost split builders can copy rather than guess at.
• Memory as a small background job. Extracting decisions with claude-haiku-4-5 after each turn and replaying them as a summary block gives an agent continuity across sessions without growing its context window.
• Enforcement has to sit outside the model. Sente Labs pairs OpenExecutive with extensible-mcp because, in the company's words, "a model can't be trusted to police itself." Every server load, tool search and call is checked in code the agent can't reach, which matters once an agent is acting with GC- or CFO-level authority. Every action is logged, and spend is capped by day, model and session.
• Self-hosting keeps company data in-house. Apache 2.0 plus ready-made Fly.io configs mean a team can run the whole system, including the RAG index built from its own internal documents, on infrastructure it controls.

What to watch

Most of the Hacker News debate is about whether any of this survives contact with a real company. Commenters are skeptical an LLM can replace the informal networking, high-stakes deal-making or investor confidence a human executive provides, even while conceding that a lot of routine executive communication is formulaic enough to automate. With 11 pull requests and 2 issues already open, the nearer-term question is whether Sente Labs keeps pace with contributions once the novelty wears off.

Key takeaways

• OpenExecutive is an Apache 2.0, self-hostable AI executive team built by Sente Labs and released on 26 August 2026.
• Eight specialist agents (CSO, CFO, Chief HR/People Officer, General Counsel, COO, CMO, CPO and a Board Communications Director) sit behind one Executive Orchestrator running claude-sonnet-4-6, with four roles switching to claude-opus-4-7 for harder decisions.
• It keeps episodic memory of past decisions through a claude-haiku-4-5 background pass into SQLite, and structures prompt caching to reach up to an 85% hit rate.
• The Hacker News launch thread reached 788 points and 538 comments, and the repository hit 1.2k stars and 71 forks within about a day.
• It ships alongside extensible-mcp, Sente Labs' separate security layer that checks every tool call in code the model can't reach.

FAQ

What is OpenExecutive?

OpenExecutive is an open-source, self-hostable AI system from Sente Labs that acts as a company's executive team. Eight specialist agents cover strategy, finance, HR, legal, operations, marketing, product and board communications, coordinated by one Executive Orchestrator so the user only sees a single consistent voice.

Which AI models does it use?

The orchestrator and most specialists run claude-sonnet-4-6. The CSO, CFO, General Counsel and Board Communications Director switch to claude-opus-4-7 with extended thinking for harder decisions, and a background claude-haiku-4-5 pass extracts episodic memory after each response.

Is OpenExecutive free to run?

The code is Apache 2.0 licensed and built to be self-hosted, with Fly.io configs included for the API, UI and QA environments. It still calls the Anthropic API directly, so a company needs its own Anthropic API key. Sente Labs describes token and dollar spend as visible and capped by day, model and session.

Why is it trending on Hacker News?

It launched directly on Hacker News as a reply to a story about a CEO firing developers to make room for AI. The submission title frames it as developers open-sourcing an AI CEO in response, and the thread has drawn 788 points and 538 comments.

Source: OpenExecutive on GitHub Discussion: Hacker News

For more autonomous business and coding agents like this one, browse the Agents directory on AI Agents Listing.