OPEN MODEL · APACHE-2.0 · RUNS ON YOUR MACHINE
Knot AI Knot AI

Two small models
that run on your machine.

Knot AI is PointBreakLab's on-device model family. Two specialists, no cloud: one writes your commit messages, the other explains what changed. Both run entirely on your hardware, no API, no code leaving your device.

KNOT SCRIBE · 7B

Writes your commits

Reads a diff, writes a correctly-typed Conventional-Commit message. Open weights on Hugging Face and Ollama.

KNOT DELTA · 9B

Explains what changed

Reads a change and writes a grounded report: what changed, why, and the risk. Built into Knot.
🦙 Get Scribe on Ollama 🤗 Hugging Face

git diff → Knot Scribe · 7B → fix: prevent division by zero

01 KNOT SCRIBE

Diff in, commit out.

Knot Scribe reads the change and returns a structured message. No chat, no prompting, the same contract it was trained and evaluated against.

01

Reads your diff

It takes the git diff, plus a little file context for big multi-file changes, packed so the model sees the whole commit's scope, not just the first file.

02

Runs on your hardware

Knot Scribe runs locally via llama.cpp or Ollama. Metal, CUDA, or CPU: same model file, same output, only speed scales. Works offline.

03

Writes the message

Structured JSON output constrained to {title, body}, a correctly-typed Conventional-Commit subject and body, with fabricated issue-refs and sign-offs stripped.

It's the model behind Knot's hands-free auto-commit, but it's open and standalone, so you can point it at any git workflow.

02 SCRIBE BENCHMARK

Measured, not claimed.

Sixty hand-reviewed commits, balanced across all eleven commit types, scored through the shipping model. First, how Scribe has improved release over release. Then a head-to-head of the current model (v5) against the previous release (v4) on this set.

TYPE ACCURACY, RELEASE OVER RELEASE

v3
40%
v4
55%
v5 · current
60%

The benchmark got harder at v5. v3 and v4 were scored on an earlier JS/TS-only set; v5 is scored on the balanced eleven-type set used everywhere below, where v4 re-scores 48.3%. So v5's real gain over v4 is +11.7 points, larger than the release numbers show.

metrics.txt · gold-v1 · 60 commits, all 11 types
Metric v4 v5
Type accuracy 48.3% 60.0%
Scope accuracy 68.3% 75.0%
Title word-overlap 27.2% 19.1%
Exact title 6.7% 0.0%
Generation errors / 60 7 4
by-commit-type.txt · correct / n · gold-v1
Type n v4 v5
feat 6 4/6 4/6
fix 6 5/6 4/6
refactor 6 0/6 2/6
perf 6 2/6 2/6
test 5 4/5 4/5
docs 6 4/6 5/6
build 5 0/5 0/5
ci 6 5/6 6/6
chore 6 5/6 6/6
revert 2 0/2 0/2

Honest read: v5 lifts type accuracy (48 to 60) and scope (68 to 75) and cuts the format errors nearly in half (7 to 4), with the biggest gains on the types v4 got wrong, refactor (0 to 2) and ci and chore up to near-perfect. Title word-overlap and exact-title fall (27 to 19, and 7% to 0%) for one reason: v5 paraphrases. It writes its own titles, often more descriptive than the single human commit it is scored against, and exact-match penalizes that. build and revert stay hard (a convention clash and too few examples). Single-reference scoring understates real usefulness: a message marked wrong for not echoing the human's exact words is often still correct.

03 KNOT DELTA

The model that explains what changed.

Where Scribe writes the commit, Knot Delta reads it: a 9B model that turns a change into a grounded report of what moved, why, and the risk. It runs on your machine, built into Knot and open below. Every number below is measured against real git history, beside the baseline a lazy model would score.

Change-type F1 (intent read from code)

0.409

vs 0.032 majority-class

12.9× base rate · n=235

Cited files that exist in the diff

100%

never hallucinates a file

holds on unseen Go

Reports fully grounded

98.5%

325 held-out commits

first-pass 88%

Exact-line precision

0.64

0.94 lenient (right region)

honest: file + region reliable

grounding.txt · exact-line precision · 4 unseen repos
repo language verified line precision
flask Python 11/12 74%
gin Go · unseen 12/12 70%
ripgrep Rust 12/12 52%
frozen 325 JS/TS + 320/325 64%

Honest read: Delta never cites a file that isn't in the diff (100% across every repo, including Go, which is absent from its training data), and it always points at the right region. What it does not yet do reliably is pin the exact line: that lands 52 to 74% of the time, lowest on Rust. It knows which file and which part changed; the precise line is still improving. One capability, measured straight, with the parts that do not yet work said out loud.

04 TRAINING & FILES

Built in the open, on one GPU.

Knot Scribe is a LoRA fine-tune of a permissively-licensed base, trained only on public commit history. No user data, no private code, ever.

knot-scribe-model.txt
Base model → A permissively-licensed open code model · Apache-2.0
Method → QLoRA (4-bit) · completion-only SFT (loss on the message only)
Hardware → A single NVIDIA RTX 3060 (12 GB)
Data → ~3,227 diff→message pairs mined from permissive OSS · JS/TS, Python, Rust, Go
Privacy → No user data · no private code · public commit history only
Formats → LoRA adapter (PEFT safetensors) + LoRA GGUF (llama.cpp / Ollama)
License → Apache-2.0 (open-base derivative)

Every file (adapter weights, GGUF, tokenizer, and the full model card) is public on Hugging Face.

05 GET IT

Run it in one line.

Both models are open weights, on Ollama and Hugging Face. Point Scribe at a staged diff for a commit message; run Delta on a diff for a change report.

KNOT SCRIBE · commit message 
ollama run pointbreaklab/knot-scribe "$(git diff --staged)"
KNOT DELTA · change report 
ollama run pointbreaklab/knot-delta "$(git diff)"

Prefer llama.cpp? Each model's GGUF, files, and exact command are on its Hugging Face model card, linked below.

🤗 Knot Scribe on Hugging Face 🤗 Knot Delta on Hugging Face

Knot Scribe writes every commit message inside Knot, and Knot Delta powers its change reports. Knot is PointBreakLab's peer-to-peer version control system.