AEGIS Is Open Source
By Arshad Ansari
Three months ago I wrote about AEGIS — my flow-first personal AI orchestration system — and ended with a section titled Why It's Not Open Source Yet. The honest version was: it was wired too tightly into my own life. Secrets in the wrong places, homelab paths hardcoded, an infrastructure personality named after an anime villain who happened to own all my SSH scripts.
That's fixed now. AEGIS is on GitHub, MIT-licensed: github.com/hikmahtech/aegis.
This isn't me shipping another agent framework — there are plenty, and most of them are better funded. AEGIS is a smaller, stranger bet: that software can learn the shape of one person's life well enough to interrupt less. Open-sourcing it doesn't change that thesis. You still fork it and make it yours. What changed is that "make it yours" is now a configuration exercise instead of a fork-and-rewrite.
What "shippable" actually meant
The gap between works for me and someone else can run this is mostly boring plumbing, which is exactly why I'd been avoiding it for three months. Concretely, four things had to happen.
The hardcoded assumptions had to become configuration. AEGIS used to know that a ticket prefixed a certain way meant "route this to the infrastructure agent." That knowledge was a line of code, which meant it was everyone's line of code. Now content routing is an admin-editable table of match rules — regex, prefix, or contains, mapping to labels, assignees, and approval gates. My personal ticketing convention is one row in my database, not an assumption baked into yours.
Agent behavior moved out of Python and into data. The four personalities used to be hardcoded branches. Now their tools, routing keywords, model tiers, and chat aliases live in the database and are edited from an admin page. You can rename them, re-scope them, or add your own without touching code. They're permission boundaries with names — and now the names and boundaries are yours to set.
Secrets left the codebase entirely. Every integration — Google, GitHub, the social channels, the LLM proxy — is bring-your-own-credentials, entered through the admin UI and stored encrypted. There's no maintainer's app quietly committed in a config file that you're hoping still works. You bring your own, because that's the only honest way to ship this.
The admin panel ships inside the image. You bring up the stack, open the panel, and configure everything from there — agents, integrations, schedules, routing. You don't edit seed files to get started. And then, the least glamorous part of all: a sanitization sweep to get my name, my paths, my homelab's internal DNS, and my personal shortcuts out of the defaults, so a fresh install starts empty instead of starting as me.
What it is, for anyone new
AEGIS is a small fleet of named agents running scheduled and event-driven Temporal workflows over your own data — tasks, email, money, knowledge, homelab alerts — and asking you for a decision only when they actually need one. It's local-LLM-first through a LiteLLM proxy, reaching for a hosted model like Claude only when a job wants more horsepower. The spine is FastAPI, Postgres, and Temporal, running on a few machines at home rather than in a cloud region.
The one idea I'm proudest of is the interactions primitive. Every time the system needs a human, it's the same shape: a row in Postgres, a card in your chat app, and a workflow quietly waiting on your tap. Approvals, choices, drafts to review, plain acknowledgements — all one mechanism. That single decision is what turned AEGIS from a notification machine into a queue of interruptions that have to earn their way in.
What it is not
It is not a SaaS. There's no hosted version, no sign-up, no multi-tenant anything. It's not a chat UI you log into, and it's not a plugin marketplace. It will do nothing useful until you point it at your own accounts and your own models. The value isn't a product you switch on — it's a set of patterns you can run on your own hardware: durable workflows, human-approval gates, personality-scoped tools, local-first inference, and personal knowledge as infrastructure. If that sounds like more setup than you want, it honestly might be, and that's a fine reason to just read the code instead.
One small thing I like
The post that may have brought you here was scheduled by AEGIS itself. Writing the copy was a task in my to-do list; the labels picked the channels, the due time was when it went out, and nothing published until I approved a card in chat. It routes through a self-hosted Postiz instance that owns the actual platform connections. Which means the first real job the open-source version did was announce its own release — either poetic or a conflict of interest, and I still haven't decided which.
Why bother
I'm not open-sourcing AEGIS because it's finished, or because it'll win a benchmark. I'm doing it because the patterns travel even when my specific life stack doesn't. If you've ever wanted a careful backstage system instead of a louder assistant — one that notices the boring thing, prepares the next step, and asks for your attention only when it's actually useful — the code is now yours to take apart.
Less command center. More trusted household system. Now with a git history you can read.
Building something data-heavy? Let's talk.