holaOS
A local-first workspace for running multiple agents with shared memory, tools, and apps.
Per-dimension scores and reasoning
Evidence: README claims local-first and local storage, but no permission model details; install script via curl pipe poses supply chain risk; dependency lock exists but no vulnerability scan evidence; external effects (e.g., sending emails, publishing content) lack explicit user confirmation mechanism; rollback exists in SDK tests as reversible actions but not at product level; source attribution: publisher identity unverified. Deductions: least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, source attribution are only partially or insufficiently evidenced.
Evidence: README and CI config are consistent; test files show compile-time and runtime validation in SDK, but full test suite results not provided; dependency lock exists but availability not verified; failure messages appear in tests (e.g., invalid_state, not_connected) but not all scenarios covered. Deductions: self-consistency, dependency availability, failure messages meet basic level but not fully proven.
Evidence: README describes multiple use cases (desktop, open source, enterprise); capability boundaries explained (e.g., BYOK, MCP); trigger precision not detailed; environment fit (macOS, Windows, Linux) declared. Deductions: audience and scenarios, capability boundaries, environment fit are adequate, but trigger precision evidence insufficient.
Evidence: README structure clear; install notes detailed; naming consistency good; examples and FAQ partially present; known limitations not explicitly listed; license is modified Apache 2.0; versioning changelog not provided; maintenance responsibility has contact. Deductions: information architecture, install notes, examples and FAQ, license, maintenance responsibility are adequate, but naming stability, known limitations, versioning changelog evidence insufficient.
Evidence: Output usability (e.g., generating .xlsx, .pptx files) claimed; marginal value (multi-agent shared memory) explained; cost-benefit (built-in models or BYOK) described. Deductions: output usability, marginal value, cost-benefit are based on claims, no actual run verification.
Evidence: Claims in README partially supported by code and tests (e.g., SDK tests), but cross-source corroboration insufficient (single repository); fact-inference separation mostly clear. Deductions: claim traceability, fact-inference separation adequate, but cross-source corroboration insufficient.
- Install script via curl pipe poses supply chain risk; review script content before execution.
- Publisher identity unverified; proceed with caution.
- License is modified Apache 2.0 with additional commercial conditions; review before commercial use.
- Known limitations and changelog not provided; may affect long-term maintenance.
What does this agent do, and when should you use it?
holaOS is an Electron desktop workspace that can run Claude Code, Codex, and its built-in holaOS agent side by side. It keeps context, preferences, and project history in locally stored, readable, editable plain files shared across those agents. Its HolaApps are interactive application surfaces opened beside the agent, and a user can define a HolaApp with a URL and an MCP server. The project describes both built-in models and bring-your-own-key paths for OpenAI, Anthropic, and compatible endpoints; local desktop development is managed through npm runtime-preparation and launch commands.
A user can run Claude Code, Codex, or the holaOS agent in one workspace and give them the same memory, skills, integrations, and apps. Memory is stored as local plain files; agents can act through connected integrations, MCP servers, browsers, and files. HolaApps open as live interactive surfaces inside the workspace; the README names Notion and a browser as examples and allows a user-defined HolaApp to combine any URL with an MCP server. The project also states that it can save deliverables as .xlsx, .pptx, and .docx files, receive tasks from Feishu, WeChat, Slack, and Telegram and return results to the originating thread, and run scheduled or trigger-based automation.
- A developer who alternates between Claude Code and Codex but wants both to retain one project context and set of tool connections.
- An individual workspace user who needs an agent to work across local files, a signed-in browser, and connected business tools.
- An operator who wants to watch and take over work occurring in an interactive Notion or browser surface beside the agent.
- A team member who needs agent output saved as a usable .xlsx, .pptx, or .docx file rather than left as chat text.
- A user who wants to submit work from a Feishu, WeChat, Slack, or Telegram thread and receive the result in that same thread.
What are this agent's strengths and limitations?
- Places Claude Code, Codex, and a built-in agent in one workspace with stated shared memory, skills, tools, and apps.
- Stores memory as local plain files that users can inspect and edit.
- Provides HolaApps as live, side-by-side interactive surfaces and supports custom URL-plus-MCP-server apps.
- Offers both built-in models and BYOK paths for OpenAI, Anthropic, and compatible endpoints.
- Desktop development and launch depend on Git, Node.js, npm, and Electron; the documented installer targets macOS, Linux, or WSL.
- Using a personal model provider requires maintaining apps/desktop/.env and the corresponding keys or endpoint settings.
- Browser, OAuth integration, messaging-channel, and MCP use require network access plus relevant third-party accounts or servers.
- The supplied licensing evidence needs review: repository metadata says NOASSERTION, while the README describes a Modified Apache 2.0 license with additional commercial-distribution and branding conditions.
How do you install or deploy this agent?
For macOS, Linux, or WSL, run:
curl -fsSL https://raw.githubusercontent.com/holaboss-ai/holaOS/refs/heads/main/scripts/install.sh | bash -s -- --launchFor the documented manual path, first verify git, node, and npm. Then run from the repository root:
npm run desktop:install
cp apps/desktop/.env.example apps/desktop/.env
npm run desktop:prepare-runtime:local
npm run desktop:typecheck
npm run desktop:devThe README says the installer installs Git and Node.js 24.14.1/npm when missing. apps/desktop/.env is the recommended environment-file location; provider settings or keys are needed only when using a user-supplied provider.
How do you use this agent?
After installation, run npm run desktop:dev to launch the Electron desktop app. Choose Claude Code, Codex, or the built-in holaOS agent and use the shared memory, tools, and apps in the workspace. When changing local runtime code, run npm run desktop:prepare-runtime:local before launch; to fetch the published runtime, run npm run desktop:prepare-runtime. Use the built-in models or configure your own OpenAI, Anthropic, or compatible endpoint credentials.
How does this agent compare with similar options?
Compared with running Claude Code or Codex separately, holaOS positions them and its built-in agent as sharing one workspace, memory, tool set, skills, and apps. The supplied material provides no feature or performance benchmark against those alternatives.