Wonda CLI
Generate, edit, and publish AI images, video, music, and social content from your terminal — your coding agent already knows how to drive it.
- Source repo
- degausai/wonda (no longer public)
- Stars (last seen)
- ★ 162
- Last updated
- 23d ago
- Primary language
- TypeScript
- FA score
- 44/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificCodex · Claude Code
- Cost
- Needs a paid subscription or licence
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- An indie e-commerce seller asks Claude Code to turn one product photo into a sora2 product video, add music and animated captions, then publish it to TikTok in one pipeline.
- Not a fit if
- Teams that need fully offline, local-only media processing
- Individuals unwilling to create a wonda.sh account or buy credits
- Teams needing self-hosting or source modification (proprietary licence)
- Source review
- 44/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
Wonda is a proprietary command-line tool that brings AI content generation to the terminal, covering images, video, music, speech, and direct social publishing. Every command outputs JSON to stdout, making it trivially consumable by shell scripts and AI coding agents such as Claude Code and Codex; it also ships skill files and native plugins so agents learn every command and workflow automatically. Generation jobs run in the wonda.sh cloud on a credit basis, while publishing connects directly to TikTok, Instagram, X, Reddit, and LinkedIn. Editing targets TikTok/Reels-style short-form video with more than 20 operations including animatedCaptions, splitScreen, lipsync, background removal, and upscaling. Installation is available via npm, Homebrew, and native agent plugins.
Wonda organizes capabilities into subcommands: generate image|video|text|music creates content from text prompts or reference images (example models include nano-banana-2, sora2, and suno-music); audio speech|transcribe|dialogue handles speech; edit video --operation ... offers editing ops such as animatedCaptions, textOverlay, merge, splitScreen, trim, speed, skipSilence, birefnet-bg-removal, topaz-video-upscale, and sync-lipsync-v2-pro; analyze video produces a composite frame grid plus an audio transcript for video understanding; publish instagram|tiktok (with carousel variants) posts directly; the linkedin, x, and reddit command families cover search, browsing, posting, likes, and messaging; scrape social|ads pulls competitor profiles and Meta Ads Library data; and blueprint manages reusable workflows. All output is JSON with built-in --jq field selection; --quiet returns only job IDs for piping, and JSON mode engages automatically when stdout is piped.
- An indie e-commerce seller asks Claude Code to turn one product photo into a sora2 product video, add music and animated captions, then publish it to TikTok in one pipeline.
- A social media manager batch-produces Reels-style clips: merging segments, burning animated captions, removing silence, and adjusting speed.
- A marketing team uses
scrape adsto search the Meta Ads Library for competitor ad creatives. - A personal brand operator manages LinkedIn and X posting, replies, likes, and DMs directly from the CLI.
- A podcast/short-form creator uses transcribe, enhanceAudio, and voiceExtractor to clean up and extract audio tracks.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npm i -g @degausai/wondanpx skills add degausai/wonda
- Install notes cover
- macOS
How do you install or deploy this agent?
Install via npm or Homebrew:
bash
npm i -g @degausai/wondaor:
bash
brew tap degausai/tap && brew install wondaThen authenticate and install the agent skill file (a wonda.sh account is required):
bash
wonda auth loginwonda skill install -o .
Plugin installs for AI agents:
bash
npx skills add degausai/wondaClaude Code:
bash
/plugin marketplace add degausai/wonda
/plugin install wonda@degausaiGemini CLI:
bash
gemini extensions install https://github.com/degausai/wondaHow do you use this agent?
Authenticate and top up first (account at wonda.sh; generation costs credits):
bash
wonda auth login
wonda topup
wonda balanceGenerate an image:
bash
wonda generate image \
--model nano-banana-2 \
--prompt "Product photo of headphones on marble" \
--wait -o photo.pngGenerate a video from a reference image:
bash
MEDIA=$(wonda media upload ./product.jpg --quiet)
wonda generate video --model sora2 \
--prompt "Slow orbit, dramatic lighting" \
--attach "$MEDIA" --duration 8 --wait -o video.mp4Add TikTok-style animated captions:
bash
wonda edit video --operation animatedCaptions --media "$VID_MEDIA" \
--params '{"fontFamily":"TikTok Sans","position":"bottom-center","highlightColor":"#FFD700"}' \
--wait -o captioned.mp4Publish to Instagram:
bash
wonda publish instagram \
--media med_abc123 \
--account ig_acct_456 \
--caption "New drop. Link in bio."Extract fields with built-in jq:
bash
wonda generate image --model nano-banana-2 --prompt "A cat" --wait \
--jq '.outputs[0].media.url'What are this agent's strengths and limitations?
- Native agent integration: the skill file auto-syncs, so an agent reading --help can independently handle model selection and full workflows with no human learning curve.
- All-JSON output with built-in --jq and a --quiet mode makes it naturally scriptable for multi-step content pipelines (generate → music → captions → publish).
- A closed loop from generation through editing to direct publishing on TikTok/Instagram/X/Reddit/LinkedIn, plus competitor ad scraping and analytics.
- Proprietary licence with hard dependency on the wonda.sh cloud: generation consumes credits and cannot be self-hosted or run with local models.
- Requires an account and prepaid credits before any use; cost structure must be estimated via
wonda pricing estimate. - Social publishing and LinkedIn/X/Reddit actions rely on stored session credentials, adding credential-management and platform-compliance considerations.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Wonda CLI This agent | 44 · Major gaps | CLIPaid | ★ 162 | 23d ago | TypeScript | Codex · Claude Code |
| Publora Social Publishing Skills | 66 · Some gaps | Agent plugin / skillFreemium | ★ 70 | 2d ago | Python | Codex · Claude Code · Claude.ai |
| md2wechat: AI-Agent CLI for WeChat Official Account Publishing | 70 · Some gaps | CLIFreemium | ★ 3.8k | 1mo ago | Go | Codex · Claude Code |
| SuperCMO Creative Skills | 71 · Some gaps | Agent plugin / skillFree + model costs | ★ 324 | 9d ago | Python | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The materials describe local and remote modes, API key and OAuth authentication, and explicitly say agents must not handle login credentials. However, the product supports broad account actions including reading, publishing, deleting, liking, and messaging; confirmation requirements are limited, and least privilege and rollback are not well explained. Resources and tool code show reads of balances, capabilities, and account skills, plus HTTPS service calls. Local credential handling such as the stated Keychain storage helps, but dependency security, source attribution, and consistent safeguards for external actions are thin.
The README, skill file, and MCP code broadly align on generation, social platform actions, and local/remote modes. The skill sample cuts off during its role description, and the files do not fully align in coverage of install channels and integrations. API failures have JSON responses or MCP error codes, and the README says errors go to stderr, but systematic troubleshooting, dependency availability, and recovery evidence are limited.
The product targets terminal users and AI assistants that can run shell commands, and lists generation, editing, research, and publishing scenarios. MCP instructions describe local/remote boundaries, account selection, and relay-offline behavior in useful detail. However, the skill description is broad and trigger conditions and boundaries are not specified for every capability. The README claims macOS, Linux, and Windows support, while the skill describes some desktop features as Mac-specific, giving partial environment guidance.
The README provides installation instructions, command categories, examples, output formats, pricing, and a license statement; plugin and Gemini metadata add integration details. Deductions reflect differences between the README command list and skill coverage, plugin version metadata without a changelog, and limited information about known limitations or maintenance responsibility. The license is identified as proprietary, but its terms are deferred to an external site.
JSON output, quiet mode, field selection, and built-in jq suit scripting and agent use. Examples show an end-to-end generation, editing, and publishing workflow, and pricing estimates and balance commands support cost awareness. The product covers many content and social tasks, but usage costs credits and some actions directly publish or change account state; cost control and pre-publication review guidance are thin.
The README, skill file, plugin metadata, and MCP source provide some corroboration for product scope, integrations, and operating modes. Many capability and safety statements remain documentation claims without source-level evidence for each claim. Version, change, and maintenance information is limited, and the distinction between verifiable facts and promotional claims is incomplete.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Generation consumes account credits. Publishing, messaging, liking, following, or deleting can affect real accounts; verify the account, content, and destination before invoking these actions.
- Routing between the local relay and cloud twin depends on mode and availability. Some desktop features described in the skill are Mac-specific, and the supplied skill sample is truncated during its role description.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
WAB prepares spare browser copies while the Mac is locked and reuses them for new personas. Until a spare is ready, a new persona's first launch still needs a locked screen. Anonymous screenshots, recordings and brand extraction use temporary headed WAB sessions. When WAB_PROXY_URL or WAB_PROXY_SERVER is set, anonymous capture refuses the standalone headless path instead of bypassing that proxy. Brand images and fonts use the headed browser context's request client. …View on GitHub ↗ -
v1.72.6 Wonda 1.72.6
Release notes excerpt
Wonda 1.72.6 improves commands in hidden browser windows on a busy Mac. It disables Chrome's background renderer deprioritization while preserving normal idle page-timer throttling. Set WAB_REUSE_ONLY=1 to reuse a healthy matching driver and refuse implicit browser starts or restarts. The show/focus guard can revalidate a persona's last known browser PIDs when a full process scan times out. These changes need CLI and driver 1.72.6. Update the runtime through the normal WAB updater when personas are safely stopped, coordinating existing sessions with their owners. …View on GitHub ↗ -
v1.72.5 Wonda 1.72.5
Release notes excerpt
Wonda 1.72.5 fixes browser commands that kept running after timing out or losing their client. Queued commands now respect deadlines, cancelled input releases its keys/buttons, closing a stuck tab cancels its work, and late tab setup cannot replace a newer tab. Focus fallback and tab-title reads are bounded. Snapshots wake background content with a lingering focus lease; get stays read-only. The release also includes personal invite links with nullable limits and the authenticated download path. …View on GitHub ↗
FAQ
Does Wonda cost money?
wonda topup, check wonda balance, and pre-estimate costs with wonda pricing estimate.Does my AI agent need extra configuration?
wonda skill install, the skill file syncs in the background; the agent reads --help and the built-in skill to discover commands, models, and workflows on its own.Which platforms are supported?
What credentials are needed to publish to social platforms?
auth set and verified with auth check.Can I chain commands into a pipeline?
--quiet prints only IDs, --jq selects fields, and JSON mode activates automatically when stdout is piped. The README shows a full generate → music → captions → publish pipeline.Is Wonda CLI safe to use?
Thanks — we use this to improve the page.