Start with a document.
Point Plancast at Markdown, plain text, a text-based PDF, or a public article URL. It prepares a conversation grounded in your source, with checks before speech generation.
YOUR READING. A NEW WAY TO LISTEN.
Give your
ideas a voice.
Turn Markdown, text, PDFs, and public articles into two-person audio briefings. Hear the main ideas, evidence, and caveats in a conversation shaped by the content.
PL AN / CASTLOCAL CLI
# A document becomes a conversation.
$ plancast PLAN.md
--length 2m --play
01 / JANEThe curious voice
02 / GEORGEThe practical perspective
Source → dialogue → audioOn your Mac
Local generation by defaultTwo distinct AI voicesPause, seek & change speedNo API key needed for local use
LESS READING. MORE UNDERSTANDING.
A briefing, built
from your source.
For articles, reports, research, notes, and plans you want to understand. The content guides the conversation, regardless of file extension.
Follow the conversation.
Jane asks questions tailored to the source; George explains its ideas and evidence. Plans can include proposals and explicit next steps. Articles do not become implementation reviews. Both voices are AI-generated.
Listen on your terms.
A local browser player gives you pause, forward, rewind, and speed controls. Choose two or five minutes. Keep the audio, transcript, and source-reference sidecar.
YOUR MAC DOES THE WORK.
Local AI.
One less service.
Qwen3 writes the dialogue through a Plancast-managed llama.cpp runtime. Pocket TTS gives it two voices. No Ollama or LM Studio needed.
First setup downloads the software and models. After setup, local generation keeps your source on your Mac. Node is bundled; Python, uv, and FFmpeg are managed for you.
Explore models and voices ↗FROM TERMINAL TO BRIEFING.
Three commands.
Then press play.
Already have Homebrew? Start here. Otherwise, download the standalone archive with its bundled runtime.
Download standalone for Apple Silicon ↗v0.6.1 prerelease · macOS 14+
Intel support is unverified. The archive is not a notarized installer.
1. INSTALL
brew install sarthakdabhi/tap/plancast2. CHOOSE YOUR WRITING MODEL
plancast setup-local3. LISTEN TO YOUR SOURCE
plancast PLAN.md --length 2m --playReplace PLAN.md with your file. Setup lets you choose Qwen3 4B (2.5 GB), 8B (5.0 GB), or 14B (9.3 GB) and remembers your choice. Downloads need internet, plus additional space for speech assets and dependencies. Generation can take several minutes.
CHOOSE WHERE IT RUNS.
Local by default.
Cloud when you choose.
Use your Mac, an OpenAI API key, or a Gemini API key. Choose writing and speech providers independently, or use one provider for both.
Qwen + Pocket TTS
Jane and George voices. Once models are downloaded, generation stays on your Mac with no cloud API charges.
plancast report.pdf --play
OpenAI
GPT-4.1-mini writes the dialogue; GPT-4o-mini-TTS voices it with Alloy and Nova. Set OPENAI_API_KEY.
plancast report.pdf --provider openai --play
Google Gemini
Gemini writes the dialogue; Gemini TTS voices it with Kore and Puck. Set GEMINI_API_KEY.
plancast report.pdf --provider gemini --play
Cloud writing sends extracted source content; cloud speech sends the generated dialogue. Each selected cloud stage requires acknowledgement before uncached processing. Applicable API usage is billed to your account. Gemini TTS is a preview service; live Gemini generation and voice quality have not yet been verified in this project.
API keys, models, voices, and setup ↗A few things
to know.
Does my source leave my Mac?
Not with the default local providers. Setup downloads assets, but local generation processes your source on your Mac. If you choose cloud writing, extracted source content is sent to that provider; cloud speech sends generated dialogue. Plancast discloses each cloud stage before uncached processing. Your API account pays for that usage.
Can I change the voices or the model?
Yes. Jane and George are the defaults; Alba and Marius are also supported. Run plancast setup-local to choose Qwen3 4B, 8B, or 14B, or use plancast setup-local --model qwen3:8b. A successful setup saves your choice for future runs. Smaller models have not been benchmarked here. See the configuration guide for options.
Can I choose the conversation style?
Auto framing follows the source content. A Markdown article is still an article, and a text file can contain a plan. Use --framing document to focus on ideas and findings, or --framing plan for a proposal review. Neither option permits inventing missing facts or next actions. See the framing guide.
Can I combine sources or review a draft first?
Yes. Pass up to ten sources, choose --length 5m, and use --audience technical or --focus "risks". Use --transcript-only to review a validated draft, then plancast render draft.json to voice it unchanged. Local caches retain source snapshots and generated content until you run plancast cache clear. See the workflow guide.
How accurate are the briefings?
Plancast checks structured dialogue and source references before generating speech. These checks cannot guarantee semantic accuracy or complete coverage. This is an early prerelease; use the source document to verify decisions and important details.
Is Plancast free?
Plancast is open source under the MIT license. Local generation has no cloud API charges and uses your own hardware. Downloaded models, runtimes, and voice recordings retain their respective licenses.