MindsHub Cowork

MindsHub Cowork is the unified workspace built by MindsDB for delegating entire knowledge-work tasks to open-source AI agents. You describe an outcome in plain language, and the agent plans the work, connects to your data sources, runs code in a transparent scratchpad, and returns finished, shareable artifacts. The platform is open source under the MIT License and runs on the web, as a desktop app for macOS and Windows, or self-hosted from source.

What It Is

MindsHub Cowork sits at the intersection of an AI agent harness and a productivity workspace. Rather than offering a chat interface for back-and-forth Q&A, it accepts whole-task briefs — "build a KPI dashboard from Postgres and Stripe," "extract 38 receipts into a Q2 expense workbook," "run a daily 7am briefing across Slack, Gmail, and Linear" — and returns finished documents, dashboards, spreadsheets, or HTML apps. The platform is described by MindsHub as "the unified workspace where open-source agents get work done for you."

Architecture: How the Platform Fits Together

Cowork is the user-facing layer on top of a managed runtime with several distinct components:

• Agent Harnesses — Interchangeable open-source agents (Anton and Hermes in Cowork today; OpenClaw, NanoClaw, and OpenClaw + GBrain available standalone) that can be swapped from a dropdown without losing workspace state.
• Model Router — A single endpoint that routes each step to the right LLM by cost, latency, and capability. Supports frontier models from Anthropic (Claude), OpenAI (GPT), and Google (Gemini), plus open-weight models including DeepSeek, Qwen, and Kimi. Models are referenced via latest:* aliases that stay current as providers ship new versions.
• Credentials Vault — Encrypted secrets store scoped per connection; agents never see raw API keys or database credentials.
• Scratchpads — Protected, reproducible execution environments where every SQL query, Python script, and intermediate table is visible and replayable.
• Memory & Skills — Cross-session memory that persists schemas, preferences, and past decisions, plus a reusable skill library that the agent recalls when the work fits.

Connectors and Data Access

Cowork connects to a wide range of data sources and SaaS tools through a connectors directory. Featured integrations include PostgreSQL, Stripe, Gmail, Slack, Google Calendar, Linear, HubSpot, Zendesk, Gong, Google Drive, BigQuery, and Notion. Credentials are stored in the vault and scoped per connection, so the agent can reach live data at runtime without exposing raw keys.

Artifacts and Collaboration

Agent output in Cowork is turned into concrete artifacts — HTML dashboards, Markdown documents, Excel workbooks, and code — rather than a scrolling chat transcript. Artifacts can be kept in the workspace as drafts or published to a live, shareable URL. Work is organized into projects, and scheduled tasks let the agent run recurring jobs (daily briefings, weekly summaries, recurring audits) on a set cadence.

Open-Source Lineage and Current Status

MindsHub is built by MindsDB, founded in 2018 in Berkeley by Jorge Torres and Adam Carrigan. The company's About page states that MindsDB grew from an open-source in-database ML project into what it describes as "the most widely adopted open platform for AI analytics and data workflow automation," before expanding into LLMs and open-source agents and rebranding to MindsHub. The GitHub repository (mindsdb/minds) is released under the MIT License and, per GitHub metadata, had 39,360 stars and 6,209 forks as of mid-2026. The latest release listed on GitHub is v26.1.0, published April 23, 2026. The platform is SOC 2 Type II certified, per the MindsHub website. The company states it has raised $50M+ from investors including Benchmark, Mayfield, Y Combinator, and NVIDIA, and has a community of 800+ contributors.