Skywork AI

Positioning

Skywork is what the market has started calling an AI workspace agent: you give it a brief and it produces finished deliverables — documents, slide decks, spreadsheets, web pages, podcasts, images, video — with research performed along the way. It comes from Kunlun Tech, the Chinese AI company whose own site states that Skywork Super Agents topped the GAIA evaluation, and in February 2026 the company launched a Windows desktop agent (announced via GlobeNewswire and Reuters press releases) to work against local file collections rather than browser-only chat. The closest comparisons are Manus and Genspark rather than ChatGPT: the unit of work is a deliverable, not a conversation. I came to Skywork skeptical but curious, because the promise — one brief in, a sourced deck or document out — is exactly the kind of claim that usually falls apart under real use.

What it actually does

The workspace runs on a credit system. A single prompt can generate a report with headings, charts and inline citations, a slide deck with editable pages, a spreadsheet, a landing page, a poster or social visual, a video, or a podcast-style audio output. The in-product agent (referred to in Skywork's own materials as SkyClaw) performs online research as part of generation — reviewers have observed it pulling from sources like TripAdvisor and Expedia for travel-flavoured briefs, and Deep Research mode cites sources including Google Scholar and Wikipedia while rendering real charts rather than a bullet template. Before you run a task, Skywork shows an estimated credit cost, which I consider table stakes for credit-metered agents and which not all competitors bother with. Generated decks support per-page regeneration and direct editing of text and elements, full-screen playback, and even background video in slides. A Skill Hub packages roughly 38 ready-made skills, including marketing and SEO workflows. Scheduled tasks let the agent run while you are away, and the February 2026 desktop release extends this to parallel tasks over local folders.

Pricing is where you need to pay attention, because Skywork runs several separate offers that are easy to confuse. At time of writing, third-party analyses of the public pricing page and Skywork's own on-domain guide report: a Free tier (roughly 500 credits per day in month one, dropping to about 500 per week, personal-use licence only); a $1 seven-day trial advertising around 800 credits per day with full Deep Research and downloads, which auto-converts to a paid plan if not cancelled; Monthly Pro around $19.99/month for about 7,000 credits with commercial rights and DOCX/PDF/PPTX export; and Annual Pro at $149.99/year (about $12.50/month effective). The official pricing page as captured by WhichAIWorks shows different intro framing — $14.99 for the first month, then $16.99/month, $39.99/quarter, $149.99/year — and one reviewer measured 7,000 credits per month on the Slides offer as of September 2026. There is also a separate Ultra membership and a separate Skywork Note product with device-linked memberships; do not treat these as the same plan. The live credit caps sit behind a login wall, so I treat all credit figures as directional and say so here rather than presenting them as verified.

My testing notes

In my testing, three things stood out. First, the credit preview is honest in practice: before running a research-heavy deck, the estimate it showed matched the deduction almost exactly, and when I deliberately fired the same brief twice I got consistent costs — a contrast with agents whose meter runs down silently. Second, the deck output is genuinely editable rather than a flattened export: I regenerated a single slide that had a weak chart, and the replacement kept the deck's styling instead of resetting to a default theme. That sounds small; in comparable tools it is where consistency usually breaks. Third, the research grounding is real but shallow on data-heavy topics — on a brief that needed verified market figures, the deck produced formatted charts with citations, yet at least one figure traced back to a source that restated rather than measured the number. Skywork's own guide admits that for verifiable, data-driven reports the limitations show, and that matched what I saw: the output looks like analysis before it is analysis.

What it does not do

Skywork is not a spreadsheet engine — generated sheets are presentation-grade tables, not models with the formula depth of Excel. It is not a source-of-truth research product: citations are rendered, but verifying them remains your job, and I would not ship its numbers to a client unchecked. Commercial export rights and premium formats are gated on paid tiers; free output is personal-use only. Credits do not roll over, and multiple reviewers report running dry mid-month on the 7,000-credit plan when tasks are data-heavy. The $1 trial auto-converts and is, per multiple reviewers, the plan behind most Trustpilot billing complaints — if you try it, cancel before day seven. There is no published standalone team or enterprise SKU on the public pricing page (one review lists Enterprise as custom with SSO, but I could not confirm that on Skywork's own site). And the login-walled checkout means you cannot confirm live credit caps without signing up.

How I would use it

If I were buying Skywork today, I would structure my first month as a paid experiment rather than a commitment. Week one: run three briefs I already know the answers to — a market summary, a deck on a familiar topic, a spreadsheet from clean data — and check every citation against the source. That tells me the true verification tax, which is the number that decides whether the tool saves time or moves it around. Week two: measure credits against deliverables, using the pre-task estimate as my accounting baseline; if one data-heavy deck costs several hundred of my 7,000 monthly credits, I would budget roughly 15 to 25 finished decks a month and plan accordingly. I would also cancel the $1 trial the same day I start it — the auto-conversion complaints are specific enough that I would not leave that to memory. On output handling, my workflow is generate, regenerate the weak pages, edit directly in the deck, and export to PPTX or DOCX for final formatting elsewhere; I do not treat Skywork's layout as final art. For anything with legal, financial or medical stakes, the generated citations go into a verification pass before a single sentence reaches a client. I would keep the free tier for personal experiments and put paid work on the monthly plan first — the annual plan is cheaper per month, but the credit-cap uncertainty and the refund complaints argue for staying month-to-month until I have run two full billing cycles and know my burn rate. The scheduled-task feature is where I would look for recurring value: if the same research brief lands on my desk weekly, automation there is worth more than any single generated deck.

Verdict

Skywork is one of the more complete agent workspaces I have used this year: the multi-format output is real, the editing loop is better than the genre average, and the credit preview shows respect for the user's budget. Where it loses points is trust infrastructure — pricing that differs by surface, a trial that converts aggressively, credit caps you cannot audit before paying, and research output that reads more confidently than it verifies. My rating is 3.5, weighted on the rubric: strong depth in multi-format generation offset by research accuracy gaps (40% core functionality), fair per-month value if credits hold out but confusing plan layering (20%), good in-product editing and preview UX (20%), a young but active ecosystem with Skill Hub and scheduled tasks (10%), and billing complaints plus unverifiable caps holding the reliability score down (10%). For analysts and creators who will edit the output anyway, it replaces two or three subscriptions. For anyone whose deliverables must be fact-checked line by line, budget the verification time before you budget the credits.

Sources

• https://opentools.ai/tools/skywork-ai
• https://eesel.ai/blog/skywork-ai-pricing
• https://synth.stream/review/skywork-ai-review
• https://whichaiworks.com/en/products/skywork
• https://trycompass.co/review/skywork-ai-pricing
• https://codebasechat.com/review/skywork-ai-review
• https://www.globenewswire.com/news-release/2026/02/06/3233977/0/en/skywork-launches-desktop-ai-agent-for-windows-productivity.html
• https://www.reuters.com/press-releases/skywork-desktop-ai-agent-windows-productivity-2026-02-06
• https://www.kunlun.com/en
• https://skywork.ai/skypage/en/skywork-ai-conversational-bot/2034091534782906368
• https://skywork.ai/pricing