Task Relay coordinates AI agents, local applications and multi-step project work from Telegram. It keeps track of the handoffs, files and decisions, so you can focus on the work.
Send a request in your messenger. Relay finds the right task or helps you prepare a workflow, with clear inputs, outputs and steps to review before work starts.
Agents and tools do their part. You choose an output, ask for a revision, pause or continue. Original requests, file versions and decisions stay attached to the work, even after a restart.
Relay runs on your computer. Your selected AI providers handle the work; Telegram brings back progress, files and questions. An optional Apple Messages text pilot is available on macOS.
02 / Programs
What Relay supports today.
These integrations are implemented in Task Relay. Connect your own accounts and install the tools you want to use. Recently added features need an updated Relay runtime; see the download’s version notes below.
Blender — modeling, rendering & animation
Create scenes from geometry data, inspect and revise selected models, import textures and objects, render previews, and animate transforms or cameras. Saves editable Blender files. Video encoding also needs FFmpeg and ffprobe.
Rhino 7 & 8 — modeling & rendering
Create, inspect and revise native Rhino models with reviewed Python scripts. Save in the selected Rhino version and render named views with Rhino Render. Requires Rhino and a usable license on macOS.
Create presentations with editable text, shapes, tables, charts and images. Relay saves a native .pptx file without requiring PowerPoint or a Google account. Implemented in the updated Relay runtime with presentation support installed.
Codex, Claude & Gemini — agent work
Continue Codex desktop tasks, run Codex CLI workers, use managed Claude sessions, or use Gemini file workers. Codex and Claude can edit files and run commands within their permissions; Gemini’s file-worker route writes declared text files.
OpenAI, Gemini, Qwen, DeepSeek & OpenRouter — text APIs
Run text conversations and read project files. A direct text API connection does not itself provide application control or file editing.
Gemini & Veo — images, speech & video
Use configured Gemini models for image generation/editing and speech, and Veo for video. The available output depends on your selected model and account access.
Browser research & messaging
Bounded browser tasks and an optional Perplexity account-browser pilot, with their own setup. Telegram is the main task and file-delivery interface; Apple Messages is a separate macOS text pilot.
Direct local application execution is currently supported on macOS. Windows task execution is not supported yet. Grasshopper is paused; there is no native Keynote, Word, Excel or PowerPoint application-control integration.
03 / File formats
What you can open, create and save.
These file routes are implemented in Relay. Each row names the tool that produces the file. Native models and PowerPoint presentations remain editable.
Implemented support: inputs and saved outputs
Program / task
Inputs
Creates and saves
Blender models & assets
.blend scenes/libraries; .png and .jpg textures; prepared .json geometry or reviewed .py scripts
Editable .blend, .png previews, .json checks; asset import can also save a .zip bundle
Blender animation
Selected .blend and a prepared .json animation manifest
Editable .blend, H.264 .mp4 video and .png frames/preview
Rhino 7 / 8
.3dm models; reviewed .py scripts and .json modeling/render settings
Editable .3dm in Rhino 7 or 8 format, .png previews/renders and .json checks
PowerPoint presentations
Prepared .json slide content and selected .png/.jpg images
Editable .pptx with native text, shapes, tables and charts. Requires the updated runtime with presentation support; absent from beta.1.
Gemini images
Text prompts and supported .png, .jpg or .webp references
.png, .jpg or .webp, according to the model’s returned image format
Gemini speech
Text and configured voice/model
.wav; also .mp3 when FFmpeg conversion is available
Veo video
A video prompt with a configured Veo model
.mp4
Text, data & code workers
Project text files and your instructions
Text files such as .md, .txt, .csv, .json, .py, .js, .html and .css
Documents, spreadsheets & other exports
.docx, .xlsx, .pdf and .svg need a capable Codex/Claude worker with the appropriate libraries or export tools. These are worker-dependent outputs. PowerPoint .pptx generation has its own implemented Relay operation, listed above; it does not include existing-template editing or PDF export.
Additional image providers
OpenAI and OpenRouter image-generation stages are also implemented and save PNG candidates. They need an updated Relay runtime and the relevant provider credentials/model access. These stages are not included in the beta.1 download.
Formats without a built-in route
Native Keynote .key and Grasshopper .gh/.ghx are not supported. CAD/mesh exchange formats such as .dwg, .dxf, .obj, .stl, .fbx and .glb are not standard Relay export operations; any such export needs a separately scoped tool or script.
A program’s ability to open a format does not mean Relay can automate it. Support depends on the operation, installed tools and selected files.
04 / Workflows
A starting point for the next task.
These example workflows show the direction of current development. Starter selection is not included in this downloadable beta; begin with an instruction in Telegram and review the available next steps.
Research → evidence → report
Check sources, collect evidence and turn selected findings into an editable report.
Architecture → model → presentation
Research constraints, review a concept, build in Blender or Rhino, and prepare a presentation.
Native model → revision → checked candidate
Inspect an existing model, review an exact edit, then check the candidate and its preview.
Approved carousel → reel → review
Plan an adaptation from selected cards and assets, then produce and inspect a reel with configured media tools.
Data → analysis → charts → presentation
Validate a dataset, make reproducible calculations and build an editable presentation from the results.
Starters describe the steps and required tools. Each stage needs its own agreed scope; choosing a starter does not start the whole workflow or install missing integrations. These starters are part of current development.
05 / Get started
Make room for the work.
Download the Mac beta and follow the setup in the app. Prefer a terminal? The CLI is available below.
Package contents: beta.1 was built before dedicated PowerPoint generation and the OpenAI/OpenRouter image stages were added. Those features are implemented, but require an updated runtime; the downloads below do not include them yet.
Install. Open the DMG, drag Task Relay into Applications, then eject the disk.
Connect. Open the app and follow its saved setup steps: AI access, Telegram bot, background service and pairing.
Start a conversation. Send your first instruction in Telegram. The companion stays in the menu bar for settings and controls.
Early beta: this build has no Apple Developer ID or notarization. After trying to open it, macOS may require Privacy & Security → Open Anyway. Only approve the download if you trust it. Apple’s opening instructions. Updates may require refreshing permissions; managed Macs may block this beta.
Python is included. You’ll need your own AI access and a dedicated Telegram bot; provider usage may cost money. First-time Messages setup remains a separate pilot. Optional tools such as Blender are installed separately.
For developers on macOS or Linux. Requires Python 3.11+, curl and tar. Download this beta’s source, then run the guided terminal setup:
curl -fLO https://task-relay-website-production.up.railway.app/downloads/Task-Relay-0.12.1-beta.1-source.tar.gz
tar -xzf Task-Relay-0.12.1-beta.1-source.tar.gz
cd task-relay-0.12.1-beta.1
sh install.sh --terminal-setup
The installer creates its own environment and prints the installed CLI path. Keep that folder in place. Setup saves connections without starting provider work. Follow the printed instructions to start Relay and pair Telegram.
The CLI honors the data connection already selected in the companion. Existing login services are preserved; switching their owner requires a reviewed handoff. Linux service support depends on an available user service manager.