A local AI coding agent for Windows — no cloud required

v0.2.0  ·  CPU  ·  Windows x64

What is Ultima Agent

Ultima is a terminal-based AI coding agent that runs entirely on your Windows PC. It spins up a local llama.cpp inference server, gives the model a full tool belt, and drops you into an interactive REPL where you can ask it to read files, write code, run shell commands, search the web, and remember things across sessions — all without sending a single token to the cloud.

Point it at any directory and it becomes an autonomous coding partner that works inside your project, not in a sandboxed cloud environment.

[#]
100% Local
Runs llama.cpp on CPU. No API keys, no subscriptions, no telemetry.
[T]
Real Tool Use
Read, write, search files. Run PowerShell. Diff git. Fetch URLs.
[M]
Persistent Memory
Durable knowledge index survives across sessions. Learns as you work.
[W]
Web Search
Optional Tavily integration for real-time web lookups, cached locally.

Download

Latest Release
UltimaLlama-CPU-Windows-x64.zip
Download for Windows x64
CPU build  ·  Windows 10 / 11  ·  x64

Or install in one line from an elevated PowerShell window:

PowerShell irm https://ultima.globalwarningnetworks.com/install.ps1 | iex

Installation

The one-liner above handles everything automatically. For a manual install:

Prerequisites

Step 1 — Download and extract the package

PowerShell # Download Invoke-WebRequest https://ultima.globalwarningnetworks.com/downloads/UltimaLlama-CPU-Windows-x64.zip ` -OutFile "$env:TEMP\UltimaLlama.zip" # Extract all binaries to ~/.ultima/bin New-Item -ItemType Directory -Force "$env:USERPROFILE\.ultima\bin" Expand-Archive "$env:TEMP\UltimaLlama.zip" -DestinationPath "$env:TEMP\UltimaLlama-extract" Copy-Item "$env:TEMP\UltimaLlama-extract\UltimaLlama\bin\*" ` "$env:USERPROFILE\.ultima\bin\" -Force

Step 2 — Add to PATH

PowerShell [Environment]::SetEnvironmentVariable( "PATH", "$env:USERPROFILE\.ultima\bin;$env:PATH", "User" ) # Restart your terminal after this

Step 3 — Put your GGUF model in C:\models\

PowerShell New-Item -ItemType Directory -Force "C:\models" # Copy GLM-4.7-Flash-Q4_K_M.gguf (or any .gguf) here

Step 4 — Run

Open a new terminal, navigate to the folder you want to work in, and type:

Terminal cd C:\my-project ultima

Ultima starts llama-server.exe in the background, loads your model from C:\models\, and drops you into an interactive agent session. Type /help to see available commands.

Recommended Model

★ Best Results

Run GLM-4.7-Flash (Q4_K_M) for the best experience with Ultima Agent. GLM-4.7-Flash's strong instruction following, tool-use accuracy, and long-context reasoning make it the ideal match for Ultima's agent loop. File: GLM-4.7-Flash-Q4_K_M.gguf  ·  ~17.5 GB

Download GGUF models from HuggingFace and place them in C:\models\. Ultima scans that directory recursively and lists every .gguf file at startup. The installer can download the model automatically when you run it.

Context size is chosen automatically based on model file size:

Model sizeContext window
20 GiB +4 096 tokens
15 – 20 GiB8 192 tokens
10 – 15 GiB16 384 tokens
Under 10 GiB32 768 tokens

Your last model selection is remembered between sessions.

What Ultima does best

Every write and shell operation requires your explicit y / N approval before anything happens. Read-only operations (file reads, searches, git diff) run directly without prompting.

REPL Commands

CommandDescription
/helpShow available commands
/clearClear conversation history (keeps system prompt)
/knowledge [query]Search the knowledge index, or list document count
/reindexRebuild the knowledge index from disk
/remember <text>Save a note directly to the knowledge base
/web <query>Run a Tavily web search (requires TAVILY_API_KEY)
/fetch <url>Fetch and read a web page as plain text
/quit  /exit  /qExit Ultima (session is saved automatically)

Environment variables

VariableDefaultPurpose
ULTIMA_HOME%USERPROFILE%\.ultimaHome directory for binaries, knowledge, sessions
TAVILY_API_KEY(unset)Enables web search via Tavily