Help & Knowledge Base

IKKF is a free, open-source, local-first framework that turns a single prompt into a structured 7-phase reasoning workflow (Intent → Retrieval → Decomposition → Reasoning → Verification → Composition → Confidence), backed by a knowledge base of atomic, verifiable facts. It makes smaller/cheaper AI models perform like domain experts on specific domains — no fine-tuning required.

Yes — IKKF is free forever. No license, no key, no machine-lock, no expiry. There is no capability gating, ever. You can support the project on a pay-what-you-want basis (☕ $5 / 🍺 $15 / 🚀 $50) if it saves you time or money, but that's entirely optional and doesn't unlock anything.

Python 3 (3.9+) and an AI provider. IKKF works with any OpenAI-compatible provider — including a free local model via ollama (recommended, no API cost) or a cloud provider like OpenAI. If you have Python and an internet connection, you can run it.

On macOS/Linux, run:

curl -fsSL https://ikkf.info/install.sh | bash

On Windows, download and run install.ps1 from the homepage, or use the manual ikkf-cli.zip install tab.

This is usually a Python version mismatch. IKKF needs Python 3.9+. Check your Python version:

python3 --version

If it's older than 3.9, install Python 3.9+ and re-run the installer. The installer creates a self-contained virtual environment (~/.local/share/ikkf-cli) so it won't interfere with your system Python.

Ensure pip is available for your Python:

python3 -m pip --version

If it's missing, bootstrap it with python3 -m ensurepip (or install via your package manager), then re-run the installer.

If a download fails, check your internet connection and that your network allows access to ikkf.info. The installers verify a SHA256 checksum of the CLI zip — if you see a checksum error, re-download (it may have been corrupted mid-transfer). Both install.sh and install.ps1 pin the correct checksum, so a mismatch means the zip didn't download cleanly.

The installer writes an ikkf wrapper to ~/.local/bin. Make sure that directory is on your PATH. Open a new terminal and check:

echo $PATH | grep -o "$HOME/.local/bin" || echo "~/.local/bin is MISSING from PATH"

If it's missing, add it. On macOS/Linux (add to your ~/.zshrc or ~/.bashrc):

export PATH="$HOME/.local/bin:$PATH"

Then open a new terminal and run ikkf --version.

This is the most common issue. If you previously had ikkf installed (for example via pip install), the old entry point may take priority over the new wrapper because of PATH ordering. To find out which version is actually running:

which ikkf && ikkf --version

If it points somewhere other than ~/.local/bin/ikkf, the old install is shadowing it. Uninstall the old copy, for example:

pip uninstall ikkf

Then remove any stale entry point and open a fresh terminal so the new wrapper takes priority.

The installer adds the wrapper location to PATH, but PowerShell may need a restart (or a new terminal window) to pick it up. If it still isn't recognised, manually add %LOCALAPPDATA%\Python\Scripts (and the wrapper's folder) to your system PATH via System Properties → Environment Variables, restart your terminal, and run ikkf --version.

Run ikkf init to set up your provider. You can use a local model via Ollama (free, no API key) or any OpenAI-compatible provider. For Ollama:

ikkf init

then select the Ollama option and pick a model (a 7B-class model works well).

Run:

ikkf start "Build a REST API in Python"

IKKF decomposes the goal and drives all 7 phases automatically. You'll see structured output as each phase completes.

Create atomic knowledge files — one concept per file, each with a source — and point IKKF at them. A good start is ~20 concept files for a domain you know well. IKKF then retrieves from this base and verifies every claim against a source, so you can audit exactly why it gave any answer.

The core is local-first and runs on your machine. If you use a local model via Ollama, you can run fully offline. If you use a cloud provider (e.g. OpenAI), you need an internet connection for the model calls.

With a local model, everything stays on your machine. With a cloud provider, your prompts go to that provider as normal for any API use. IKKF itself doesn't phone home — there's no license check and no telemetry requirement.

The project is still in its early release phase and the repository isn't published yet. It's open in spirit and free — a public source release is on the roadmap.