Built for SMEs. Watch It Work.

Most AI coverage is written for companies with an AI team. Small and mid-market firms don't have one, and they're the ones losing the most hours to work a machine could do: the proposal that takes an afternoon, the Friday report that eats half a day, the file review that eats a morning.

Odokai is built for that gap. Not a chatbox, and not another subscription stacked on the seven you already pay for. One agentic platform where every person in your firm gets a team of AI agents behind them, built once, shared by everyone, improving with every use, and running inside your own environment. Your team, multiplied.

Rather than describe it again, we've put two walkthroughs up. They're screen recordings of the real product, recorded in one take. Models pause mid-run, a demo lands flatter than expected, and we left that in. If you're evaluating a platform you intend to put client work through, the unedited version is the one worth watching.

1. What the platform actually is

The first video is a tour: logging in, finding your way around, and watching an agent do a piece of work while you see what it's doing.

A few things in it are worth calling out, because they're the difference between a chat window and a place work gets done.

You pick an agent, not a model

Every session starts by choosing which agent you're working with. The suggested prompts change to match it, and the model is already set for that agent, though you can override it. Your team builds their own agents, each with its own skills, tools and access. Nobody has to learn which model is good at what.

You can see what the agent is doing

While a request runs, the context panel lists the tools the agent is calling and the files it writes. When it went off to research renewable energy market trends in the video, the working steps were visible the whole time, not hidden behind a spinner. That visibility is the same mechanism the audit trail is built on.

Agents and people share one file system

The workspace explorer is the part that reframes things. It's an ordinary file system, folders and all, that both your people and your agents can reach. In the video an agent is asked to put some sample documents in a folder, then to add a CSV calendar, and the files simply appear where you'd expect them. You can preview them, download them, or ask the agent to open one for you rather than clicking through yourself.

That's what makes agent output usable. The work lands as files your team already knows how to handle, in a place they can get to, instead of trapped in a chat log.

Governance is in the product, not the brochure

Permissions are set per conversation: run on auto, or have the agent ask before it uses a tool, with the defaults your administrator sets for the team. Alongside that sit approvals, execution history, schedules, and the permission levels an admin decides each agent is allowed. As the video puts it, Odokai is designed to work as part of a team, and that means a team of real people as well as a team of agents.

There's also a marketplace for installing skills and apps, integrations into the tools you already run, webhooks to trigger agents from other systems, and a command palette that will search your past sessions or spin up a new agent without leaving the keyboard. Help is context-aware: it knows which screen you're on and explains that screen.

2. Turning that into capability that compounds

The second video is the how-to, and it's the more important of the two. It shows the primitives being layered: files, tools, skills and agents stacked into something repeatable.

The task is deliberately small: get our brand styles out of the website and use them to produce on-brand documents that print cleanly. It runs on the default assistant every user gets. No specialist agent, no engineer.

1. Capture the asset once. The agent is asked to pull the styles from the Odokai site into a creative folder: a style guide, the design tokens, the CSS. That folder now exists for everyone.
2. Use it. Ask for an on-brand HTML page and the agent builds it from those saved tokens rather than guessing at your colours.
3. Package it as a skill. A skill is a bundle of instructions and capabilities behind a single command, and you create one by asking for it in plain English. In the video: a skill that generates landscape, on-brand pitch decks from the existing brand assets.
4. Run it against anything. One line invokes the new skill against a fresh brief. It gathers the content, builds the deck with click-through navigation, and offers a print view for PDF.

You can do things quickly, group them together, and get real productivity gains from instructing the agents once, saving the assets and capabilities, and then asking them to do it again.

The example is a deck. It could as easily be your tender response format, your monthly client report, your file-review checklist. The pattern is what matters: one person works out how a job is done, saves it as a skill, and everyone else runs it from then on. That's how a small team stops re-solving the same problem every week.

What it does to a week

Agents do the assembly, the drafting, the creative and the first-pass checking. The people you pay for judgement spend their day on judgement instead of admin.

• Proposals and tenders. Four hours becomes twenty minutes. A brief goes in, a branded first draft comes out with your pricing and terms already in it. Your best person reviews it instead of writing it.
• The Friday report. Half a day becomes fifteen minutes. Numbers pulled from every system, formatted, and checked against last month before anyone opens it.
• Contract and file review. All morning becomes minutes. First-pass review flags the risk clauses, pulls the key dates and figures, and lists what's missing.
• Research and background. Two days becomes an hour, with the citations attached. The decision happens this week, not next.
• Campaign pages and brand assets. Two weeks becomes an afternoon. Copy, imagery and layout, on brand, reviewed by your team and published.

A twelve-person firm starts turning work around like a hundred-person one. The big firms manage it with headcount. You won't have to.

Why this shape suits a smaller firm

The expensive part of AI is never the second workflow. It's the first: the edge cases, the integrations, the sign-offs, the governance. So that's the only one that carries a full build fee.

We put your first workflow into production in one to five weeks for a fixed fee agreed up front. That build stands up your instance, your governance and your integrations. After it lands, your team builds the next one on the same foundation without paying another rollout fee, and most clients ship their second in days. Workflows three to five follow the same pattern and cost less each time. Skills compound the same way inside the platform, which is exactly what the second video is showing.

Set that against the salary-hours your team currently loses to manual work in an ordinary week and it comes in under the junior hire you were pricing up, for the work you were going to hire them to do. It also replaces the accidental stack of overlapping subscriptions most firms have accumulated, which costs far more than the invoices suggest.

Your data doesn't leave

This is the reason most smaller firms in regulated and professional services have stayed on the sidelines, and it's designed for rather than bolted on. Your documents, your index, your audit trail and your workflow logic stay inside your environment. At the point of reasoning, a request goes to a governed model under a zero-retention, no-training agreement, hosted in-region. Where a step can run on a smaller local model instead, it does, so less ever needs to leave at all. You can hold the model provider contract directly if you'd rather.

Run it shared and hosted by us, or dedicated in your own AWS account, managed or self-hosted. Whatever you build, you keep: workflows, agents, prompts and knowledge bases export in documented formats on request, on any plan, whether or not you renew.

Start with the job your team dreads

The videos show the platform on our work. The next step shows it on yours. Bring a handful of your own documents to a free call of about forty-five minutes and we'll build an agent that handles one of your workflows, live, on your data. Not a demo script.

You'll know inside the hour whether it's worth doing. You'll leave either with written acceptance criteria and a fixed fee for the build, or with a much clearer picture of what's actually worth automating. Both are worth the hour.

Book the session, or read more about how the engagement works. More walkthroughs are on the way, including working with your own files and knowledge bases.