/01
You ask
A task, in plain words, from the desktop app, the terminal or Telegram.
Self-Hosted AI Agent
It runs on your hardware, talks to your model server, and keeps memory and chats in a local folder.
or install from the terminal
Atomic Agent 
Hermes Agent
GAIA Level 1 – 53 real-world tasks on the same model and same machine
Accuracy · GAIA Level 1
Atomic Agent 
Hermes Agent +11.3pp More accurate
37/53 solved. Grammar-constrained decoding (GBNF) forces structurally-valid tool calls; bounded tails keep reasoning on-budget.
1.6× Faster per task
217s vs 351s. cache_prompt + slot_id reuse across steps, plus parallel read-only batches that overlap I/O after a single inference.
15 vs 9 Often right
Grammar-locked actions and step-by-step checks mean Atomic Agent rarely walks into an unrecoverable wrong answer.
how it works
↻ the loop repeats – KV-cache keeps each step cheap, nothing re-sent to a server
/02
Local model decides
One pass on your own GPU returns a list of next actions — grammar-locked, always valid.
/03
It runs them
Browser, files and shell — in parallel. Pauses for approval on risky steps.
/04
Repeat → done
Loops with cached context, so each step stays fast — until it replies or you stop it.
Questions or feedback? Grab 15 minutes with the team, any slot works.
Book a callfeatures
the engineering that keeps it fast, private, and running on your own machine
A byte-stable prompt prefix means every turn reuses the same cached compute. Nothing is re-encoded, and inference stays on your own hardware.
prefix ▓▓▓▓▓▓▓▓▓▓ cachedtail ░░░ re-computed────────────────────step N KV-cache HIT ✓
Grammar-constrained decoding forces structurally valid actions, even from tiny local models. How GBNF tool calling works
root ::= "[" call ("," call){0,15} "]"call ::= { "tool":…, "args":{…} }
Curated Q4 GGUF quants fit a long window on an everyday GPU.
fp16 ████████ ~24 GBq4 ██ ~6.7 GB
Recall injected in the tail, a few hundred tokens, cache intact.
SELECT content FROM memories_fts ORDER BY rank→ ~320 tokens
It pulls the backend and model for you. No accounts, no bills.
$ atomic-agent models startchat: started, healthy● on port 8080
Use cases
INV
IMG
ZIP
- Scanning ~/Downloads · 318 files
- Reading docs for real titles
- Grouping → /Invoices /Shots
- Renamed 212 · deduped 7
files, folders, done Sorts, renames, groups, and cleans your files — reading each one to understand it first.
BTC/USD64,210 ▲
NEWS llama.cpp v0.5 ships
MATCHARS 1–1 CHE
→ @you
- Cron · every 30m
- BTC $63,420 (+2.1%)
- Arsenal 2–1 · 78’
- Pinged Telegram ✓
cron in, Telegram out Schedule a recurring check — markets, news, match scores — and get a Telegram ping only when something changes.
acme/api main
cloning…
#42 NullRef on init
PR #128
- Cloning acme/api
- Reading issue #214
- Found in loader.go:88
- PR fix-214 ready ✓
issue in, PR out Clones a repo, reads an issue, finds the cause in the code, and opens a fix on a branch — pausing before it pushes.
wikiarxiv.org
PDF
note.md source: arxiv ✓ saved to memory
- Browsing 6 sources
- Parsing 2 PDFs
- Extracting · 12 cites
- Note saved · searchable ✓
browse, read, save Drives a real browser across sources, reads PDFs and pages, and writes a cited note into your local memory.
any model, local or cloud
Atomic Agent on OpenRouter Real production traffic: teams run Atomic Agent against cloud models through OpenRouter.
Total tokens 72.9M
Models used 32
Active since May 2026
Get new releases
by email
The desktop app is out for Mac, Windows and Linux. Download it here, then leave your email and we'll tell you when a new version ships.
EMAIL ADDRESS
FAQ
Self-hosted AI agent, answered.
-
What is a self-hosted AI agent?
A self-hosted AI agent is an agent whose loop, model and data run on hardware you control, instead of a vendor cloud. Atomic Agent runs on your computer or server and keeps its memory and chats in a local folder.
-
Which model servers does Atomic Agent support?
It can download and manage llama.cpp for you, or connect to your own llama-server, Ollama, LM Studio or vLLM. Cloud providers such as OpenRouter are optional.
-
Can I run it on a server without a screen?
Yes. The terminal version runs as a service with scheduled tasks, and you can talk to it from Telegram or Discord. The browser tool can run headless.
-
Does it send any data out?
With a local model, the model and your data stay on your machine. The app does send anonymous usage statistics and crash reports by default, and you can turn them off with one setting. Tools that browse or call APIs reach the network when you ask them to.
-
Is it really free?
Yes. Atomic Agent is MIT-licensed. The only costs are your hardware and electricity, plus any cloud provider you decide to connect.