Pandrator

Pandrator is a local-first workspace for creating audiobooks, subtitles, and voiceovers. It brings document preparation, transcription, optional AI correction and translation, speech generation, review, and export into one browser interface.

Local speech and transcription models can keep media on your own computer. Cloud language, translation, and speech providers are optional and used only when you configure and select them.

Install Pandrator

Open the latest release and choose the Manager for your operating system:

The Manager creates and owns one Pandrator workspace under the selected parent directory. It installs the application and only the local components you choose; Docker and WSL are not required. Start with one speech or transcription engine and add others later.

See the installation guide for checksum verification, workspace selection, headless setup, and updates.

Choose a workflow

The documentation index links the complete workflow, operations, security, reference, and development guides.

What Pandrator can do

Audiobooks

• Import TXT, PDF, EPUB, DOCX, MOBI, or pasted text.
• Clean difficult documents with deterministic tools, OCR, and an optional reviewable AI workflow.
• Detect structure and chapters, normalize text, and split it for speech.
• Generate speech in segments, compare takes, edit text, regenerate selected passages, and apply optional RVC conversion.
• Export WAV, MP3, Opus, FLAC, or M4B with chapters, metadata, and cover art.

Subtitles and voiceovers

• Start from SRT subtitles or common audio and video formats. WebVTT, ASS, and SSA uploads are recognized while the editing pipeline uses SRT as its working subtitle format.
• Transcribe speech with timestamps and optional diarization.
• Correct, translate, split, merge, and retime cues in a reviewable editor.
• Generate synchronized speech without changing the source media.
• Export subtitles, audio, or video with selectable audio and subtitle tracks.

Voices and providers

• Use local TTS services or configured cloud speech providers.
• Connect local OpenAI-compatible LLM servers and supported cloud LLMs.
• Record or upload reference voices, retain their transcripts, and preview built-in voices before generation.
• Import fine-tuned XTTS bundles and RVC models while keeping original and converted takes available for review.

The providers and voices guide explains the local engines, compute choices, voice cloning, and when data leaves the Pandrator host.

Reviewable AI, including no-extra-API dispatch

LLMs are optional. Pandrator can use a configured provider for document cleanup, subtitle correction, glossary-aware translation, research, and speech-text optimization. Each transformation creates a distinct revision or artifact; it does not silently replace the source text.

Passive dispatch provides a different route. Pandrator makes no model call: it snapshots the selected source and queues deterministic, leased subtitle, document-cleanup, or speech-text packets. The model already running in an MCP host claims one packet, returns a typed result over stable identities, and continues sequentially. The final artifact appears only after every packet is validated.

Read correction and translation for the choice between manual review, a configured LLM, DeepL, and passive dispatch. The exact cue, batch, timing, speech-block, and alignment model is in the subtitle pipeline reference.

Local-first does not mean local-only

Pandrator binds to loopback by default and is designed for one owner. A home server, VPN host, external HTTPS server, or GPU pod is possible when its data, Manager state, identity, and network boundary are deliberately preserved. Pandrator is not a public multi-user service.

The remote and headless guide covers supported topologies. The privacy and security guide explains local and cloud data flows, credentials, diagnostics, MCP access, and safe remote exposure.

Agent access with Pandrator MCP

pandrator-mcp lets an MCP-capable agent explain Pandrator, inspect one fixed installation, edit sessions safely, execute reviewed workflow plans, process passive subtitle, document-cleaning, and speech-text batches, import files from approved named roots, resolve live TTS catalog choices, export explicit output variants, download verified artifacts, and perform bounded Manager recovery. Target origins, absolute filesystem paths, and credentials are process configuration, not model-selected tool arguments.

On a compatible local Manager installation, Pandrator exposes an authenticated loopback Streamable HTTP MCP automatically. Stdio remains available for older hosts, source development, and the recommended remote topology where the MCP process runs beside the agent and connects to a fixed HTTPS Pandrator target.

The component guide documents:

• local, LAN/VPN, external HTTPS, and pod targets;
• least-privilege application and Manager scopes;
• owner-approved enrollment, identity pinning, rotation, and revocation;
• managed local HTTP and secret-free stdio host configuration for Codex, Claude Code, OpenCode, and Antigravity;
• passive subtitle correction/translation, PDF/EPUB source cleanup, and speech-text optimization;
• resumable local/remote source transfer and verified output delivery;
• live provider/model/voice selection and typed output variants; and
• diagnostics and optional recovery while Pandrator is stopped.

See the Pandrator MCP guide for exact installation and configuration commands.

Documentation

Public product documentation is intentionally version-agnostic where possible. Release-specific behavior and filenames belong on the corresponding release; package-specific operational contracts remain beside their packages.

Development

For source development, install Pixi and use the committed lockfile:

git clone https://github.com/lukaszliniewicz/Pandrator.git
cd Pandrator
pixi install --locked
pixi install --environment web-build --locked
pixi run --environment web-build web-build
pixi run serve-web

Run pixi run run-worker in a second terminal. See development from source for environments, test lanes, API/client regeneration, and package checks. Read contributing before preparing a change.

Getting help

Start with troubleshooting. For a bug, open a GitHub issue with your operating system, the affected model or provider, the action you took, and the smallest reproducible sequence. When an installation or service action fails, include the Manager's reviewed Download diagnostics support bundle. Always inspect diagnostics before sharing them.

License

Pandrator is released under the MIT License. This covers Pandrator's source code, not third-party dependencies, speech models, transcription models, LLMs, or voice-conversion models. Review the licence and usage terms shown before installing a model, and make sure you have the necessary rights to source media and voice material.