Tariq Hussain

Solo dev, day job by day — building Ownware nights & weekends: the open-source kit to build, own & ship your own AI agent. Any model, self-hosted.

[Tariq Hussain @TariqHu0]

Inference providers on the Hugging Face Hub is a total QoL (Quality of Life) boost

1/ After weeks of exploring model deployment options, I documented my findings in two articles. My first approach used the transformers API + FastAPI on a Hugging Face Space.

Pros: Any model, super quick deployment
Cons: Models aren’t fully optimized for production; might look into vLLM or TGI.

[Broden Suffern @brsuffern]

AI agents produce better work when you give them better context.

Snapdown captures visible text from documentation, PDFs, error messages, and tables, preserving headings, lists, code, and table structure as Markdown.

Paste it into Claude or Cursor instead of manually transcribing and reformatting it. One shortcut, entirely local.

[Tariq Hussain @TariqHu0]

@petergyang @karan4d @NousResearch I use Hermes, and I like it—but I wouldn’t call this self-learning. It’s more like an agent writing notes to its future self, then periodically cleaning those notes up. Useful? Absolutely. Learning? Only if those notes measurably improve what it does next.

[Tariq Hussain @TariqHu0]

curious about the session reuse in workflows, is that prefix caching or are you holding the whole run state in the DO?

asking bc i've been building a similar thing from the other side, self hosted instead of cloudflare native: https://t.co/76hjFvRqAg

agent is just a folder of text, one node process, any model. channels + schedules built in

[Tariq Hussain @TariqHu0]

If your CLAUDE.md has grown into a rulebook,

/doctor in Claude Code will cut it down.Ask what you can remove, not what you can add.

[Tariq Hussain @TariqHu0]

Claude Opus 5 is days old. I've had agents running on it since last night — deleting instructions, not writing them.

Anthropic cut 80% of Claude Code's system prompt for this generation. Evals held.

Your CLAUDE.md is probably a rulebook.

70s breakdown:

[Tariq Hussain @TariqHu0]

The thing I didn't expect, watching the overnight run:

conflicting instructions cost more than missing ones.

"Leave documentation as appropriate" sitting next to "DO NOT add comments" makes the model spend its thinking resolving you instead of the problem

[Tariq Hussain @TariqHu0]

@trq212 Here is video for the instructions

generated by opus 5 in 15 minutes

[Tariq Hussain @TariqHu0]

Everyone's racing to build the AI agent — the coding one, the sales one, the support one. Each welded to a single job.

I've been building the opposite: one engine that becomes whatever agent the person in front of it actually needs.

[Tariq Hussain @TariqHu0]

@JakeFleshner DIY agents.

[Tariq Hussain @TariqHu0]

@Nim_Ravid1 @sequoia @8vc @NotionHQ @DecagonAI @FarzaTV have a look, been following 'hey clicky'.

[Tariq Hussain @TariqHu0]

@16vchq https://t.co/zO3RYesMlD

[Tariq Hussain @TariqHu0]

Every screen in that clip is the real product:

· rosa/agent.json + rosa/SOUL.md — an agent is text you can edit
· ollama:llama3.2 · keyless · $0 — runs local, no API key
· the amber card — it asks before it reads/replies, scoped to #support only

· one HTTP+SSE gateway → Slack, Telegram, WhatsApp, your site, a schedule

[Tariq Hussain @TariqHu0]

An AI agent is just a folder of text.

Describe it in one sentence → watch it build itself → approve anything sensitive → it's live in Slack (any messaging door).

Keyless. Local. $0. Yours to own — not renting someone else's server.

This is Ownware, open source 👇 https://t.co/76hjFvQSKI

#AIagents #opensource #buildinpublic

[Tariq Hussain @TariqHu0]

@JakeFleshner DIY agents.

[Tariq Hussain @TariqHu0]

Build it. Own it. Ship it — into your product, to your clients, branded as yours. The one thing you can't do with a rented assistant.

Open-sourced solo. A ⭐ genuinely helps a one-person project get seen: https://t.co/poOfrXZIj8

[Tariq Hussain @TariqHu0]

Stop renting your AI agent.

I'm a solo dev — I just open-sourced the kit to build (and own) your own. Any model, your machine, from a folder of text.

Watch it go from text file to live in 30 seconds 👇

#buildinpublic

[Tariq Hussain @TariqHu0]

Ownware is the kit to build yours:

① describe it in a folder of text
② one command → a live service
③ alive in Slack, your site, your app

Support agent, shop assistant, ops bot — same kit, different text. A real engine, any model, your box.