▸ GETTING STARTED

Installing Cognithor

Install Cognithor on macOS, Linux, or Windows. Download the binary, install a local LLM via Ollama, and run the first-time setup.

Cognithor runs fully on your own machine. Installing it takes three steps: grab the binary, install a local LLM, run the first-time setup. Budget about 15 minutes for the whole thing.

System requirements

Resource Minimum Recommended
OS macOS 13+, Ubuntu 22.04+, Windows 11 Same
CPU 8 cores, 3.0 GHz 12+ cores
RAM 16 GB 32 GB
Disk 20 GB free 50 GB+ (for model weights)
GPU Optional NVIDIA RTX 3060 12GB or Apple Silicon

The 27B default planner model wants about 18 GB of RAM loaded. On a 16 GB machine, use the 7B planner and accept slightly lower quality.

Step 1: Install Ollama

Ollama is how Cognithor talks to local LLMs. It is the default backend — LM Studio and llama.cpp also work if you already run them.

macOS / Linux:

curl -fsSL https://ollama.com/install.sh | sh

Windows: Download the installer from ollama.com/download and run it.

Verify:

ollama --version

Step 2: Pull the default model

Cognithor defaults to qwen3:27b for planning. On a 16 GB machine, use qwen3:7b instead.

# Default (32 GB+ machines)
ollama pull qwen3:27b

# Small-RAM fallback
ollama pull qwen3:7b

The model downloads run 15–40 GB depending on which one you pick. The first pull is slow; every subsequent run is local.

Step 3: Install Cognithor

Download the binary from the releases page for your platform. Unzip it into a directory on your PATH (for example ~/.local/bin/ on macOS/Linux, or C:\Program Files\Cognithor\ on Windows).

Verify:

cognithor --version

Step 4: Initial setup

Run the bootstrap command. It creates the config directory, initializes the vault, and walks you through picking a model.

cognithor init

The bootstrap:

  • Creates ~/.jarvis/ with an empty vault, memory tiers, and an audit log file
  • Probes your Ollama installation to see which models are available
  • Writes a default ~/.jarvis/config.yaml you can edit later
  • Generates a per-session API token for the REST interface

When it finishes, the CLI prints a short "next steps" block. You are ready for the first run.

Troubleshooting

ollama: command not found after the install script — close and reopen your terminal, or source your shell rc file. The install script adds Ollama to your PATH but does not reload the current shell.

Bootstrap fails on Windows with a path error — Cognithor assumes %USERPROFILE% is writable. If you have redirected it via group policy, set COGNITHOR_HOME to a writable directory before running cognithor init.

Model pull hangs at 99% — the final percent is the checksum verification; it is slow on spinning disks. Wait it out.

cognithor: ollama unreachable at localhost:11434 — make sure Ollama is running (ollama serve or the system service). On Linux the installer registers a systemd unit; on macOS and Windows the installer runs Ollama at login.

What's next