A Confusing Opening
If you saw holaOS's description on GitHub, your first reaction might be "yet another AI desktop":
> Run any agent — Claude Code, Codex — across your tools (100+ integrations + MCP), apps, browser, and files, with shared memory.
But reading closely reveals an unusual positioning: it doesn't replace any agent — it lets multiple agents share the same working environment.
This is a completely different path from editors like Cursor and Windsurf that ship their own agents.
The Core Problem: Agent "Memory Silos"
If you use Claude Code and Codex on the same work, have you hit this problem:
- Claude Code refactored a module for you and has your whole project structure in context
- Switch to Codex to write another feature — it knows nothing about the refactor
- You have to manually tell Codex what Claude Code did, or have it re-read the code
- Debuggable: when an agent misbehaves, you can inspect its memory files to find out why
- Portable: memory isn't locked in when you switch tools
- Trustworthy: users have full visibility into "what the AI remembered"
- Built-in models: one account for all SOTA models (Kimi K3, GLM 5.2, GPT 5.6, Claude Opus 5, Fable 5), no API key management needed
- BYOK: use your own OpenAI/Anthropic keys, running on your own accounts
- Any MCP-compatible tool can be called by agents inside holaOS
- An MCP server you write works simultaneously for Claude Code, Codex, and the holaOS Agent
- macro: team collaboration, integrating communication, docs, CRM, and other business tools
- holaOS: developers, integrating agents, MCP tools, and local files
- Repo: holaboss-ai/holaOS
- Today: +380 stars
- Platforms: macOS, Windows, Linux
- Tech stack: TypeScript + Electron
- License: Modified Apache 2.0
- Website: holaos.ai
That's the "agent memory silo" problem. Each agent has its own context window, session history, and tool configuration. Switching agents is like swapping in an amnesiac coworker.
holaOS aims to solve exactly this. Its core promise:
> Context, preferences, and project history live in a single shared memory — stored locally, as plain files you can read and edit.
The keywords are shared and locally. Shared means usable across agents; local means not locked into a cloud service.
Three Design Decisions Worth Unpacking
1. Agent-Neutral, No User Lock-in
holaOS doesn't build its own agent; it runs Claude Code, Codex, and its bundled holaOS Agent side by side. The README says it plainly:
> No lock-in — bring the agent you already trust.
This aligns with the principle that "specialization beats unification" — holaOS provides workspace infrastructure, agents do the reasoning. You won't need to rebuild your workflow when switching agents.
2. Memory Is Files, Not a Database
holaOS stores shared memory in local files that users can open and edit directly. This design decision matters:
This contrasts with OpenAI's Memory feature — stored in the cloud, where users only see summaries and can't edit raw data.
3. BYOK Plus Built-in Models
holaOS offers two modes:
This dual-track design is pragmatic — zero friction for newcomers, control retained for power users.
Relationship with MCP
The README mentions "100+ integrations + MCP". MCP (Model Context Protocol) is Anthropic's standardized agent-tool interface protocol. holaOS supporting MCP means:
Again, "specialization beats unification" — holaOS doesn't reinvent tool protocols; it uses MCP, which is becoming the standard, to connect existing tools.
A Question Worth Asking
holaOS's positioning is clear, but one question deserves scrutiny: what is the format of the shared memory?
The README says "stored locally, as plain files you can read and edit" but doesn't specify the format. If it's custom JSON, migration across tools still requires conversion. If it's a standard format (say, Markdown + frontmatter), portability would be much stronger.
The answer determines whether holaOS is "a product" or "an ecosystem" — the former locks users into holaOS; the latter could make it the de facto standard for agent workspaces.
Comparison with macro-inc/macro
Also trending the same day is macro-inc/macro (+1180 stars), positioned as "unified workspace for teams: email, chat, docs, tasks, agents, calls, and CRM". Both build "unified workspaces", but in different directions:
macro wants to replace Notion + Slack + Salesforce; holaOS wants to replace Cursor + Claude Desktop + various MCP hosts. Different markets, but pointing to the same trend: tool fragmentation has reached a point users can no longer tolerate.
The Data
*holaOS's core bet: users don't want to rebuild their work environment for every agent. If that bet pays off, "agent-neutral + shared memory" could become a foundational design pattern for AI workspaces.*