Overview
Wordless is a local-first desktop agent workspace for macOS and Windows. Describe the work in natural language; it plans steps, calls tools, and returns something you can check. You bring the model. Sessions stay on this device.
Capabilities
Section titled “Capabilities”- Natural language: state the goal, constraints, and how you will accept the result. You do not have to pre-split the work into tool commands.
- Plan and execute: the agent breaks the job down, uses tools, and adjusts from intermediate results until there is a verifiable output. High-impact steps such as writes and commands can still wait for your confirmation.
- Multimodal work: documents, code, presentations, spreadsheets, data analysis, and image or video in one session, not a text answer only.
- Roles and reuse: stable duties become digital employees or teams; repeatable work becomes a task or a local schedule, instead of starting from a blank prompt every time.
When to use it
Section titled “When to use it”- Document generation: briefs, memos, proposals, and editable drafts
- Data analysis: cleaning, aggregation, charts, and conclusions you can check
- Decks and reports: outline to a previewable, editable presentation
- Deep research: multi-dimension lookup, sources, and evidence gaps
- Code editing: read a repo, apply a focused change, verify with tests or a diff
- Spreadsheets: formulas, pivots, charts, and reports
- Image and video: generate or revise visual assets for the task
Versus WorkBuddy and Codex
Section titled “Versus WorkBuddy and Codex”All three can run an agent. They are aimed at different jobs:
| Wordless | WorkBuddy | Codex | |
|---|---|---|---|
| Shape | Local desktop workspace | Tencent office agent workbench, with cloud account and credits | OpenAI coding agent (CLI / cloud sandbox) |
| Models | Provider or gateway you configure | Product catalog and quota | OpenAI / ChatGPT stack |
| Output | Office files, sheets, decks, and code in one session | Everyday office work with a cloud task loop | Repos, commands, and code review |
| Data | Sessions and files stay local; sync is opt-in and excludes sessions | Remote trigger; tasks can continue in the cloud after the client closes | Cloud runs happen in an isolated environment, not your desktop thread |
| Control | Tool approval and access level in the session | Product account and permission model | Codex sandbox and repo workflow |
Choose Wordless when files should stay on the device, you pick the model, and PPT / sheets need in-place edit and review. Choose WorkBuddy for Tencent office workflows and cloud continuation. Choose Codex when the work is repository coding on the OpenAI path.
Reading order
Section titled “Reading order”Finish this path before opening the rest of the product:
- Quick start: install the package that matches your Mac chip or Windows architecture from the official download page, then open the workbench.
- Configure a model: add provider credentials. On a custom gateway, use Fetch models to pull the catalog, then enable the models that should appear in Composer.
- Your first task: choose a workspace, reference a file with
@, and verify the result. - Manage conversations: find, rename, or delete local session records in the sidebar.
- Changelog: read the installed version and update notes in the app.
After the first task verifies, open Manage tasks, Image generation, workflows, automation, digital employees, and Skills / MCP by deliverable. Start with Core concepts when you need the shared vocabulary.
Open-source credits
Section titled “Open-source credits”The agent kernel comes from Pi Agent Harness, docs from Astro Starlight, the desktop shell from Electron, and the homepage 3D scene from Three.js. Preserve each project’s license and upstream link when redistributing.