LogMotif
macOS beta Apple Silicon No account Traces stay local

A local debugger for AI-agent workflows on macOS

Change one step. Re-run it. Compare both branches.All on your machine.

Record an agent run, find the step that failed, and fork it with a different instruction or model. LogMotif replays the rest, compares the two branches, and gives a heuristic improved / regressed signal — while keeping your traces on your Mac.

Built for solo developers and small teams debugging multi-step Python agent workflows.

Download free beta for Apple Silicon See the loop

v0.2.0 · signed & notarized · no account required

Python attach with live fork/re-execute · TypeScript recording preview · Offline mock mode

LogMotif and its traces stay local. Live model and tool calls still use the providers you configure (or mock mode / Ollama).

Illustration with sample data — not a live screenshot. Watch the real screen recording ↓

Getting started in three steps

1

Download LogMotif

Free macOS beta for Apple Silicon. No account.

2

Run the mock workflow

New Run works offline — no API keys required.

3

Attach a Python agent

When ready, stream your own agent into the same debugger.

LogMotif is a local development debugger — not a hosted production-observability or team-collaboration platform.

Demo

Watch the debugging loop

Failed run → pick the critic step → fork with a tweak → compare → verdict. Screen recording of the desktop app (~3.5 min) · mock mode · no API keys.

Watch the debugging loop (~3 min)

If the recording doesn’t load, try the interactive mock above or download the macOS beta and run the workflow yourself.

Screen recording · mock mode · no API keys required

In short: open a failed research run, select the critic model call, fork with a stricter instruction, compare branches, and read the improved / regressed signal.

Download free beta for Apple Silicon Send feedback

How it works

Turn a failed run into a reproducible experiment

Orchestrating agents is getting easier. Understanding why a run failed still isn’t. LogMotif is a local flight recorder for that loop.

01

Record

Model calls, tool calls, and state updates land on a timeline you can scrub — from in-app workflows or an attached agent.

02

Fork

Select the step that broke. Fork with an instruction tweak or model change. Everything before that step is kept; the rest re-runs on a new branch.

03

Replay tools from tape

On fork, recorded tool results can replay so you don’t re-hit search/HTTP APIs. Model steps after the fork still run live.

04

Compare & decide

Side-by-side branch compare with a heuristic verdict: improved, regressed, unchanged, or inconclusive — a signal to iterate, not a formal proof.

What you get

Concrete capabilities in the desktop app

⏱

Inspectable timeline

Live events, open any payload, checkpoints, and a scrubber to move through the run.

⑂

Fork and re-execute

Branch from a selected step with a prompt or model change. Compare original vs fork side by side.

◎

Tool replay (VCR)

Reuse recorded tool results on fork so experiments stay stable and don’t spam external APIs.

⇄

Exportable tapes

JSON export/import and self-contained HTML tapes for demos and bug reports — no LogMotif account.

Python SDK

Your agent. Our debugger.

Keep writing agents in Python. Stream events to LogMotif on 127.0.0.1:7741. Fail-safe by default — if the app isn’t running, your agent still runs.

Works with: plain Python · LangGraph via wrapped node spans · custom agents that emit events · TypeScript recording (preview)

Install the LogMotif attach SDK (Python 3.10+, no dependencies): 

pip install https://www.logmotif.com/downloads/logmotif-0.2.0.tar.gz

Requires LogMotif v0.2.0 or later — attach is authenticated, and the app supplies the token automatically. Open LogMotif → Attach for copy-paste snippets wired to your local port.

Download free beta for Apple Silicon Send feedback
python · attach 
import logmotif

@logmotif.record(name="my-agent", topic="Research")
def run(prompt: str):
    with logmotif.span("researcher"):
        logmotif.log_llm(
            prompt=prompt,
            content="…",
            model="gpt-4o-mini",
            provider="openai",
        )
    return "done"

# With LogMotif open → run appears → fork → compare

Why local

How LogMotif is designed differently

Hosted tracers help teams ship agents at scale. LogMotif is for builders who want the experiment loop — record, fork, compare — without a SaaS account for the debugger itself.

What LogMotif does and does not do, as of v0.2.0.
Design choice LogMotif (today)
Where run data lives Your disk (SQLite) — not a vendor cloud for traces
Debugger account None required
Core loop Record → fork → tool replay → compare → heuristic verdict
Authoring In-app workflows + Python attach (fork re-execute)
Not aiming to be Production monitoring, auto-instrumentation of every framework, or multi-tenant cloud dashboards

Download

macOS beta — free, no account

Developer ID signed and notarized. Open the DMG, drag LogMotif to Applications, run a mock workflow (no API keys).

Requires an Apple Silicon Mac and macOS 11 or later. Intel Macs are not supported in this build.

Download free beta for Apple Silicon Send feedback
Installation
  1. Open the DMG and drag LogMotif to Applications.
  2. Launch LogMotif from Applications (or Spotlight).
  3. Mock mode is on by default — click New Run with no API keys.
  4. The build is signed and notarized; macOS should not require Gatekeeper workarounds on Apple Silicon.
Known limitations (beta)
  • Apple Silicon only — no Intel or universal binary yet.
  • Python attach has full fork re-execute; TypeScript is record-focused.
  • LangGraph needs explicit node spans — not automatic instrumentation.
  • No crash reporting or product analytics in the desktop app.
  • Verdicts are heuristic signals, not formal proofs.
Verify the download

After downloading, confirm the file matches the build published here: 

shasum -a 256 ~/Downloads/LogMotif_0.2.0_aarch64.dmg

Expected:

60e58a6a8abae0087b12041027dbe75ca3f03677244175fcb533c4fecf053ad7

You can also download the checksum file and run shasum -a 256 -c against it from the same folder as the DMG.

v0.2.0 · released 31 July 2026 · arm64 · ~5.2 MB · signed & notarized · Privacy · Terms

Beta support: hello@catalistclub.com — use the feedback template; replies are founder-led.