Ghost

▄████ ██░ ██ ▒█████ ██████ ████████ ██▒ ▀█▒▓██░ ██▒▒██▒ ██▒ ██ ▒ ██ ▒██░▄▄▄░▒██▀▀██░▒██░ ██▒ ▓██▄ ██ ░▓█ ██▓░▓█ ░██ ▒██ ██░ ▒ ██▒ ██ ░▒▓███▀▒░▓█▒░██▓░ ████▓▒░▒██████▒▒ ██ ░▒ ▒ ▒ ░░▒░▒░ ▒░▒░▒░ ▒ ▒▓▒ ▒ ░ ░ ░ ░ ▒ ░▒░ ░ ░ ▒ ▒░ ░ ░▒ ░ ░ ░ ░ ░ ░ ░░ ░░ ░ ░ ▒ ░ ░ ░ ░ ░ ░ ░ ░ ░ ░

One prompt enters the wire. One local response comes back.

Ghost is a local-first terminal AI companion powered by Ollama.

The project is being rebuilt from first principals as a TUI-first cyberdeck: local inference, precise terminal behavior, and a cyberpunk interface designed for the glow of the command line.

Boot Sequence

Before opening the uplink, install:

• Go 1.26.5
• Ollama, running locally
• An Ollama chat model

Ghost currently expects the Ollama API at: http://localhost:11434/api

Load a Construct

Pull the small model used by the documented examples:

ollama pull qwen3:0.6b

Confirm that it is available in local model storage:

ollama ls

Open the uplink

From the repository root:

go run . --model qwen3:0.6b "Reply with one short sentence confirming Ghost is online."

On a successful connection, Ghost prints one model-generated response to standard output followed by a newline.

The exact response varies between models and invocations. The signal is nondeterministic; the pathway is not.

System Diagnostics

Ghost has two validation circuits: isolated repository checks and an explicit live connection test.

Offline checks

These commands do not require Ollama or network access:

go build ./...
go test ./...
golangci-lint run

Live uplink check

The tagged end-to-end test connects to a real local Ollama instance. Ollama must be running, and the selected model must already be installed

go test -tags=e2e -v .

To validate another installed model:

GHOST_E2E_MODEL=<model-name> go test -tags=e2e -v .

The end-to-end test is intentionally excluded from the normal test suite and continuous integration. Nothing reaches the live wire unless explicitly requested.

Signal Recovery

Ollama node unavailable

Start the Ollama server:

ollama serve

When Ghost cannot establish a connection, the error chain begins with:

ghost // uplink failure: request Ollama chat response:
send Ollama chat request:

The remaining text comes from the operating system and may differ between platforms.

Requested model missing

Inspect the models available in local storage:

ollama ls

Pull the requested model when necessary:

ollama pull <model-name>

A missing-model failure includes:

ghost // uplink failure: request Ollama chat response:
Ollama chat request failed: 404 Not Found:

Archived Construct

The previous v3 implementation is preserved on the archive/v3 branch and at the archive-v3-2026-08-22 tag.