shortform-ai 0.1.15

Local-first CLI for short-form video and creator analysis

shortform-ai

Local-first CLI for short-form video and creator analysis.

Recommended persistent install for repeated agent use:

pipx install shortform-ai
~/.local/bin/shortform-ai doctor

If the app is already installed with pipx:

pipx upgrade shortform-ai

Homebrew remains a good system install path once a tap is wired:

brew install shortform-ai

For one-off venv tests, use Python 3.12+ explicitly:

python3.12 -m venv .venv-shortform-ai
.venv-shortform-ai/bin/python -m pip install --upgrade pip
.venv-shortform-ai/bin/python -m pip install shortform-ai yt-dlp

Local transcription is installed on first use when no OpenAI key is configured:

shortform-ai setup transcription --local

shortform-ai analyze defaults to --transcribe auto. It uses OpenAI transcription when OPENAI_API_KEY is configured; otherwise bootstrap analyze installs local faster-whisper on first use. The default local model is base, which downloads a model cache on first use. Use --transcribe-model small for better local accuracy.

Analysis requires a real transcript. If neither OpenAI transcription nor local faster-whisper is available, analyze stops with setup instructions instead of producing a transcript-free strategy pass.

Starter Prompt For Codex / Claude Code

Paste this into an agent harness after installing shortform-ai:

Use the shortform-ai skill to analyze this reel:
{instagram-url}

Run the default agent-mode workflow, apply the agent response, and summarize the
final report with the contact sheet inline. Never print Instagram session
values.

CLI Usage

For one-prompt agent workflows, use the bootstrap command after installing the package:

shortform-ai bootstrap analyze {instagram-url-or-file} --instagram-from-chrome

It installs/updates the agent skill when a compatible skill target is detected, optionally imports the user's own Chrome Instagram session, runs analysis, fetches Instagram comments when possible, and writes bootstrap.json beside the analysis artifacts. The analyzer emits progress lines during long-running steps so the assistant does not appear stuck. When comment fetch succeeds, bootstrap merges the audience summary back into the root report.md, audience.md, and comments.json.

The default --ai auto mode is agent mode, designed for Codex, Claude Code, or another assistant interface. The CLI performs deterministic media work and writes agent_request.md, agent_schema.json, and agent_response.example.json instead of shelling out to a nested agent CLI. The active assistant should read the request, inspect the contact sheet/keyframes/transcript/comments, write agent_response.json, then run:

shortform-ai agent apply {analysis-dir}

The existing --ai codex and --ai api modes remain available for users who explicitly want nested Codex CLI execution or API-key automation.

For Instagram reel URLs with local Instagram auth configured, analysis also writes safe creator, aggregate metrics, and audio attribution metadata to media.json, reel.json, and report.md, including likes, comments, plays/views, creator profile fields, and named background music/original-audio title when Instagram exposes them.

Video analysis also tries to write contact_sheet.jpg and always writes report.md, a rich Markdown overview that points at the contact sheet when available. Each analysis includes timeline.md and timeline.json, a reverse-engineered beat timeline grounded in detected shots and transcript windows. Caption-heavy reels also get onscreen_text.md; transcript.md is the audio transcript, so it may contain music lyrics instead of the visible story text.

Development usage from a source checkout:

python3.12 -m venv .venv
.venv/bin/python -m pip install --upgrade pip
.venv/bin/python -m pip install -e .
.venv/bin/shortform-ai doctor

The CLI reads a user's own local Instagram Chrome session only after an explicit shortform-ai auth instagram --from-chrome command. Session values are redacted from output and are stored in local secure storage when available.