Show IH: CloudAxis — when your AI chat gets long, important context disappears. Here's what I built instead.

Hey IH 👋

I'm Jithu, solo founder of CloudAxis.

The problem (the real one)

Chat-based AI doesn't really "forget because you closed the tab."

The pain shows up when context grows:

• You chat longer → the window fills → early instructions and details get dropped or summarized away

• The model starts missing constraints you said 40 messages ago

• Tool results pile up → compaction kicks in → something important quietly disappears

• Research, spreadsheets, browser work — all trapped in the thread instead of somewhere durable

So you end up re-pasting old outputs, re-explaining the project, and babysitting a conversation that's supposed to be doing work.

That's not an agent problem. It's a workspace problem.

What I built

CloudAxis is an isolated cloud computer for AI agents.

The idea: don't make the chat thread the system of record. Make the desktop the system of record.

• Files on disk — research, CSVs, reports, screenshots survive between runs. The agent opens ~/files/competitors-june.csv on Tuesday; it doesn't need the whole chat history in context.

• Scheduled duties — focused jobs on cron ("every Monday 6:30am, update pricing sheet, Slack me the diff"). Each run can start lean instead of dragging 200 messages of noise.

• Real cloud browser — sessions, logins, and navigation that aren't just "paste the HTML into chat"

• Multi-agent handoffs — Researcher writes a file → Analyst reads it. Artifact handoff, not copy-paste between threads.

• Cloudia — describe outcomes in plain English; she wires specialists, skills, and duties

• Hosted models — no API keys; hard monthly caps

One line: chat is where you steer. The cloud computer is where the work lives.

Why I cared personally

I logged one week of my work. ~a third was recurring stuff that didn't need my judgment — competitor checks, Friday client reports, pre-call research, mention monitoring, social drafts.

Chat tools could do pieces of it. But every long session turned into: wait, did it remember the 23 URLs? Did it keep last week's prices? Why is it re-asking me things I said on Tuesday?

Moving outputs to files + scheduled duties fixed the reliability more than a smarter model did.

Five things I automated (~10 hrs/week back):

1. Competitor pricing — 90 min every Monday → duty + spreadsheet in Files

2. Client reports — 3 hrs every Friday → Analyst duty, PDF to workspace

3. Pre-call research — brief in Files before I wake up

4. Brand mentions — WhatsApp alerts instead of manual searches

5. Social drafts — Sunday review instead of Sunday writing

Setup: one weekend, mostly plain English. Not zero errors — but caught fast, and way better than re-explaining context every Monday.

Where we are

• Live: https://cloudaxis.ai (free tier, no card)

• Solo, bootstrapped, building in public

• Revenue: early ( $19)

Stack: React OS shell, Node/Express, SQLite, OpenClaw, Browserbase, Hetzner.

Questions for IH

1. Do you hit context loss more than "session reset"? What breaks first for you?

2. What would you put on disk vs in chat if you had a persistent agent desktop?

3. Tried Manus / Lindy / Relevance — did long-running work hold up?

I'll reply to everything. Real use cases welcome — happy to extend trials.

Deeper walkthrough: https://cloudaxis.ai/blog/tasks-i-stopped-doing-manually/
Why files beat copy-paste: https://cloudaxis.ai/blog/a-file-system-for-your-ai-agents-why-persistent-workspace-beats-copy-paste/

- Jithu