← all repositories
holaboss-ai/holaOS

An OS that actually remembers your standup updates

A local-first desktop agent that ingests your Slack, Jira, and GitHub, then compresses the mess into queryable memory you can edit.

5.5k stars TypeScript AgentsLLMOps · Eval
holaOS
Velocity · 7d
+70
★ / day
Trend
steady
star history

What it does

HolaOS is an Electron desktop app that runs a local-first AI agent connected to your work tools — Gmail, Slack, Linear, GitHub, Jira, HubSpot, and others via one-click OAuth. It auto-fetches activity, compresses it into Markdown-based memory stored locally with SQLite vec embeddings, and retrieves context through RAG so the agent doesn’t start from zero each session. You chat with it through a GUI, manage files in a visible workspace, and can inspect or edit what it remembers.

The interesting bit

The session compression is the part that isn’t just marketing. HolaOS keeps ~70% of the model context window free for fresh reasoning, preserves the active working set verbatim, and folds older history into structured checkpoints — goals, constraints, decisions, next steps — that only merge back if the live session still matches. It’s an explicit attempt to solve the slow drift into bloated, expensive long-running agent sessions.

Key highlights

  • Local-first memory: Markdown files + SQLite vec embeddings, browsable and editable by you, not locked in a cloud black box
  • 100+ integrations: One-click OAuth to workplace tools; auto-fetches signals every 30 minutes (per the comparison table)
  • Session context compression: Structured checkpoints with safe compaction to preserve continuity without token bloat
  • Subagent orchestration: Hidden parallel subagents coordinate through a single user-facing manager
  • One-account model access: Uses “leading SOTA models” without managing separate API keys (specific providers not named in README)

Caveats

  • macOS only for now; Windows and Linux are “in progress” per the platform badge
  • The “1000+ integrations” claim in the comparison table conflicts with “100+ Integration” in the feature list — the README is inconsistent on scale
  • “Modified Apache 2.0” license; the modification terms are not specified in the README

Verdict

Worth a look if you’re tired of re-explaining project context to AI assistants every morning. Skip it if you need cross-platform support today or want a headless/terminal-first workflow — this is deliberately a GUI-centric, desktop-only experience.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.