The agent that remembers: one continuous conversation, never a session.

How it remembers

A day of agent work produces far more material than conversation: tool calls, command output, file contents, reasoning. The messages you and Briglia actually exchanged are a small fraction of it, and they hold everything that matters: what you asked, what you decided, what you prefer, what was left open.

So Briglia keeps them apart. The work is pruned once it's done. The conversation has its own fractal compaction: recent messages stay detailed, older stretches are condensed into summaries, and those into summaries of summaries, always in chronological order. The result is a clear timeline of everything you and Briglia have said, from the first day to now, plus a profile of you distilled from it. It takes up to about 50k tokens of context on every request, and it pays for itself.

Most harnesses search blind: they query a memory store and hope the right fragment comes back. Briglia already knows what was said and roughly when, so when it needs an exact detail it goes straight to it in the full transcript on disk, like a sniper. Fewer searches, and the ones it does hit.

That's why it brings up a decision from weeks ago without being asked. How Briglia remembers →

What it is

Briglia is a personal assistant and coding agent with full access to the computer it runs on, in the same family as Hermes Agent and OpenClaw. It has a real shell, the file system and the network. It writes and runs code, reads and produces documents, researches the web, manages email and calendar, watches things and tells you when they change, and runs long jobs while you're away.

It acts without approval prompts. That's why we recommend installing it on a separate machine, or in a virtual machine, rather than on the computer you use every day. Give it only the keys you're comfortable giving, and read what that means before you install.

Briglia CLI is something I made for myself and chose to share. Nothing is sold here: the source is public on GitHub, and this site only explains it and gives you the install command. If, having read the code and these pages, you want it — install it.

Pick your path

How it stays yours

• Your hardware, your account. Briglia runs where you put it; everything it knows lives in your home directory as plain files you can open with a text editor, and /deleteuserdata removes it all.
• Your model. OpenCode Go, OpenRouter, any OpenAI-compatible endpoint, or a local server on your network — switch mid-conversation with /provider.
• Your line in. Only your paired Telegram chat, your terminal or your phone app can talk to it. Messages you send while it works are authenticated so a web page or an email cannot impersonate you.
• Your software. Source-available, signed releases you can verify yourself, no telemetry, no accounts, no subscription to us — you pay your providers directly.

Where to go next

• First setup — what the wizard asks and which keys you need.
• How Briglia remembers — one conversation, archived and summarised; what the model sees on every request.
• Telegram — steer Briglia from your phone, anywhere.
• Reminders & watchers — the autonomous side: schedules, conditions, external triggers.
• Commands — every briglia subcommand and chat /command.
• Security & signed releases — how updates are authenticated, and how to verify one yourself.
• License — free for individuals (commercial use included); companies need a license.