holaOS Ships Beta 0.1 Giving AI Agents Persistent Memory Across Workstreams
holaOS Beta 0.1 ships persistent multi-workspace agents with shared memory, sub-agents, and 100+ integrations — open-source and free to start
Takeaways
- Beta 0.1 ships: holaOS launches its first product release with Multi Workspaces, Sub Agents, and a Dashboard for managing parallel AI workstreams.
- Persistent memory: Agent context, preferences, and history are stored locally as plain files — survives session restarts and agent switches.
- Any agent, one workspace: Claude Code, Codex, and the built-in holaOS agent share the same memory, tools, and integrations with no lock-in.
- 100+ integrations + MCP: One-click OAuth for Gmail, Notion, Slack, GitHub, Linear, and more; plug in any MCP server for custom tooling.
- Free to start, open-source: Desktop app is free with built-in frontier models; self-host under modified Apache 2.0 with your own API keys. GitHub repo
- macOS-first for now: Signed desktop build targets Apple Silicon today; Windows and Linux support is in active development.
Most AI agents are built for a single conversation. You ask, they answer, and the next session starts from scratch. holaOS is betting that agent value only compounds when they can pick up where they left off across days, weeks, and multiple parallel workstreams. Beta 0.1 is the first release that makes that concrete.
The persistence problem
Every Claude Code session starts clean. That works for a 15-minute task. It breaks down when you're running a multi-step research agent, an automated refactor across a large codebase, or any workflow where the agent needs to accumulate knowledge between runs.
The root issue is that most agent setups have no persistent operating environment. The agent executes, terminates, and the next invocation knows nothing about the previous one. You can work around this with custom scripts, checkpointing logic, and memory stores bolted on manually, or you can use a tool designed for exactly this.
holaOS gives agents a structured operating system built from runtime, memory, tools, apps, and durable state so they can work continuously, evolve over time, and stay inspectable across runs.
What shipped in Beta 0.1
Beta 0.1 adds a workstream management layer on top of the Agent Computer foundation. Three capabilities define this version:
- Multi Workspaces: Create isolated workspaces for distinct tasks such as competitive tracking or content creation. Each workspace preserves its own context, files, and previous outputs. A content pipeline shares no state with an inbox assistant.