G2

Luca P.

Chief Operations Officer DEQUA Studio | Formerly CTO in MarTech

Marketing and Advertising

Mid-Market (51-1000 emp.)

6/11/2026

"The first file platform where my clients and my AI agents work in the same place"

4.5/5

What do you like best about Fast.io?

The workspace model is the thing that made Fastio stick for me. Each client gets their own workspace, scoped cleanly from everything else, and guests do not count against my seat allocation. I run a studio where the people touching files in any given month might be two employees, four freelancers, and a dozen client-side reviewers, and on every other platform that headcount turns into a billing conversation. Here the guests are unlimited, the plan includes a generous block of seats, and an extra internal user is one dollar a month. I stopped doing seat math entirely, which is not something I expected to write about a storage product.

Connecting my own AI agent to the workspace is where Fastio stops being a Dropbox alternative and becomes something else. I plugged Claude in through the MCP server in a few minutes, and from that point the agent reads and writes real files in the same folders my team uses. Not a sandbox, not a copy I have to keep in sync. When I ask the agent to draft a delivery summary or reorganize a campaign folder, the result lands in the workspace where a human picks it up. The scoped tokens matter more than they sound: the agent only sees the workspace and folders I grant, and every single action it takes shows up in the audit log next to the human activity. I can show a client exactly what an agent did to their files and when. That audit trail is what makes me comfortable letting agents touch client material at all.

Ripley, the built-in agent, covers the cases where connecting an external model would be overkill. It is already there in every workspace with nothing to set up, and for quick questions against a folder of documents it does the job. I use my own Claude connection for anything substantial, but having a default agent that less technical collaborators can just talk to lowers the barrier for the rest of the team.

Branded portals are the feature my clients actually see, and they have changed how deliveries feel. The share pages carry my custom URL, my background, my link preview, and my social links, so a final delivery looks like it comes from my studio rather than from a generic file host. Pair that with the share controls and the setup gets genuinely professional: passwords on sensitive links, expiration dates, download limits, and email risk screening on the people coming in. I set a proof link to expire after the review window closes and stop worrying about old links floating around in client inboxes for years.

The media handling is stronger than I assumed it would be from a platform this young. Video gets transcoded for adaptive streaming, so a 4GB master plays instantly in the browser and scrubs smoothly even on a hotel connection, while the original stays untouched. Audio files render as waveforms, which makes jumping past silence in a long voiceover take far less tedious. Previews cover the professional formats my work actually involves, PSD and AI files, RAW photos, even CAD drawings, all in the browser without the client needing the original software installed. Image transforms happen live on delivery, resize, crop, WebP conversion, which has quietly replaced a small image pipeline I used to maintain.

Comments are anchored where they belong. On a video, feedback pins to the exact frame. On a document or an image, it pins to the spot. The difference between "the logo around the middle feels big" in an email thread and a pin sitting on the actual logo at 00:42 is the difference between one revision round and three. Guests can comment without creating an account, which removes the single biggest source of friction in client review.

Metadata Views is the newest piece I have adopted and the one I am still finding uses for. You define a schema, and extraction agents pull structured fields out of unstructured files, documents, spreadsheets, images, video, even handwriting, into a grid you can verify and query. I pointed it at a folder of signed talent agreements and got names, dates, and terms into columns without retyping anything. The verification step is the right design choice, because I would not trust blind extraction on contracts, but checking a pre-filled grid is much faster than reading every page.

Tasks and approvals living inside the storage layer turned out to be more useful than I predicted. Tasks support dependencies and priorities, approvals attach to the actual deliverable, and the status is visible next to the file rather than in a separate project tool that drifts out of date. I would not call it a replacement for a full project management product, and for complex production schedules it is not trying to be. For the approve-this-file loop that dominates client work, though, having the request, the file, and the decision in one place removed a category of "did you see the version I sent" messages.

The developer surface deserves credit. Everything in the UI exists as a REST endpoint, a CLI command, and an MCP tool, and the parity is real rather than aspirational. The CLI installs from npm and handles upload, search, and sharing from the terminal, which is how I script recurring deliveries. SDKs for Node, Python, and Go exist, webhooks exist. For a content platform aimed partly at creative teams, the engineering depth underneath is unusual.

Migration was less painful than these moves usually are. The cloud import connects to the old provider over OAuth, you pick the folders, and the transfer runs in the background with the structure preserved. I moved an archive out of Google Drive this way and the folder tree arrived intact, already indexed and searchable. Two-way sync with write-back also exists for the transition period when part of the team is still living in the old tool.

The Mac desktop app rounds it out. Workspaces mount straight into Finder as native filesystem entries, files pin for offline use, and the activity stream shows what teammates and agents are doing across synced workspaces in real time. Watching an agent's file writes appear in Finder a moment after asking for them in chat is a small thing that still has not gotten old. Review collected by and hosted on G2.com.

What do you dislike about Fast.io?

The Windows desktop app does not exist yet. The Mac app is good, which makes the gap more noticeable, because half the people I collaborate with are on Windows and for them Fastio is browser plus CLI only. Windows support is listed as in development, and the web app is complete enough that nobody is blocked, but Finder-level integration is one of the platform's better features and an entire OS population currently does not get it. If your team is Windows-heavy, check the status of that release before committing to a sync-based workflow.

The credit system takes a while to internalize. Transfer, AI document processing, image analysis, and video conversion all draw from one monthly credit pool, and predicting what a media-heavy month will consume is not intuitive at the start. My first month on Professional I converted a large video backlog and watched the credits move much faster than I expected. To be fair, the platform warns you as you approach the limit rather than cutting you off, pay-as-you-go covers overruns, and on annual plans unused credits roll over, so the design is forgiving. It is the mental model that needs work, not the billing behavior. A clearer in-app projection of what a planned job will cost in credits would remove most of the guesswork.

Free tier retention has a real edge: links expire after 60 days and file retention is capped at 60 days too. For evaluating the product that is fine, but anyone planning to park client deliverables on the free plan should understand that it is not an archive. The limits disappear on paid plans, and at ten dollars a month the upgrade is not a hardship, but I have seen people assume free cloud storage means indefinite storage and this is a place where that assumption fails.

Some of the AI features I consider core are gated to paid plans, scoped RAG queries, shareable AI summaries, the plain-English activity log. Reasonable commercially, but it means the free tier does not fully demonstrate the agent-native pitch that makes the product distinctive. You connect an agent on free, then discover the deeper query features sit one tier up.

The integration ecosystem is thin compared to what a decade-old incumbent offers. There is no marketplace of hundreds of third-party connectors, and if your workflow depends on a specific niche tool talking natively to your storage, you will probably be building that bridge yourself through the API or MCP. The developer surface makes that practical, and the API has been solid in my use, but it is build versus install, and not every team has someone to do the building. Review collected by and hosted on G2.com.

What problems is Fast.io solving and how is that benefiting you?

The core problem is tool sprawl around client content. Before this, a single project's files were spread across a sync drive for working files, a transfer service for big deliveries, a review tool for video feedback, and email for everything that fell between. Each tool had its own permissions, its own links, its own version of the truth. Fastio collapsed that into one workspace per client, where the working files, the branded delivery, the frame-level feedback, and the approval all happen against the same copy. The benefit is fewer places for a file to be wrong, and far less of my week spent moving content between systems that do not talk to each other.

Giving AI agents a real working surface is the problem nobody else in this category is solving properly. The before-state was undignified: download files, paste contents into a chat window, copy the output back, save it somewhere, tell the team where. Every agent task carried manual freight on both ends. Now the agent operates on the workspace directly through MCP, reads the source files where they live, writes results where the team expects them, and the handoff from agent to human is just the file being there. Agent output went from something I transcribe to something I review.

Client review cycles compressed noticeably. Feedback used to arrive as prose descriptions of visual problems, in email, days late, and reconstructing what the client meant was its own task. With comments pinned to frames and regions, and guests commenting without an account wall in front of them, feedback arrives faster and arrives precise. Revision rounds dropped because the first round of notes is actually actionable.

Extracting information from accumulated documents went from a project to a task. The before-state for anything like "summarize the key terms across these forty agreements" was an afternoon of reading or an intern. Metadata Views turns it into defining a schema, running extraction, and verifying a grid. The information that used to be functionally locked inside PDFs is now queryable, and decisions that touch old documents stopped requiring archaeology.

Finding things in an archive stopped depending on my memory of folder names. Content gets indexed for search as it lands, and the scoped queries let me ask questions against a specific folder or workspace and get an answer grounded in those files. The before-state was the familiar one, opening six similarly named folders looking for the deck a client approved eight months ago, or asking a former collaborator where something ended up. Now the question goes to the workspace and the answer comes back with the files attached. It works at the level of contents, not just filenames, which is what makes it useful on an archive built by several people with several naming habits.

Cost predictability is the quieter benefit. Per-seat pricing punishes exactly the way a studio grows, in freelancers and client collaborators rather than employees. A flat plan with unlimited guests and one-dollar extra seats means adding a reviewer to a project is a permissions decision, not a procurement one. I invite people the moment it is useful instead of rationing access against a per-user fee.

Security questions from clients got easier to answer. Encryption at rest and in transit, complete audit trails, virus scanning on uploads, Geo-IP controls, and the agent actions logged alongside human ones. When a legal-adjacent client asked how I could prove who accessed their materials, I exported the log instead of writing a reassuring paragraph. For a small studio, being able to answer enterprise-flavored questions with actual records rather than promises is worth a lot. Review collected by and hosted on G2.com.