Caddi

What it is

Caddi is an automation platform for operations and admin teams, with a stated focus on law firms and finance. The company calls it "the deterministic AI automation platform for ops and admin teams." The reader it is written for is a non-technical person who owns a repetitive back-office process and wants it automated without learning a builder. Its own comparison pages say it suits ops teams in law and finance that want automations live immediately. The pages shown do not say where the company is based or when it was founded.

It is a different kind of "no-code" from a canvas or a form builder. You do not drag steps together. You show Caddi the task, and Caddi builds the automation for you. That makes it a fit for the workflow-automation shelf, with the caveat that the building is done by the vendor's system rather than by you.

What you build and how

The entry point is a screenshare or a chat. The homepage says non-technical teams "love training AI agents over a video call": you screenshare the job and talk it through, like training an expert new hire, and the agent picks up edge cases and keeps suggesting improvements. The company says it takes about 45 minutes on average to build your first agent. The homepage shows a demo of Caddi sharing a screen inside Docusign. Another comparison page puts it plainly: "You show it on a screenshare or describe it in chat, and change it in plain English as the work changes."

What comes out is not a macro. The company's Automation Anywhere comparison says Caddi uses the recording to understand the workflow, writes it as deterministic code and runs it over APIs, so it survives UI changes. It draws a contrast with RPA, which it says replays UI actions and breaks when screens change, and with Appian, which it describes as a low-code BPM platform where you model processes and assemble process apps. Caddi's pitch is that there is no app to design, no screens to build and no process app to maintain.

The company calls its agents hybrid. Code runs the work, and AI is used only where a step needs judgment. The homepage claims "99% of agent runs succeed" and sets that against an agent re-deciding every step and drifting by Friday. On maintenance, it says Caddi flags what drifted after a run and proposes a fix that you approve. On cost, it promises one predictable price with no token meter running behind the work.

The product is organized around three steps, named in the menu and on the homepage. Discover finds the work worth automating. The company says this is agentic discovery that starts from hypotheses grounded in industry expertise and validates them in your tools, in contrast to process mining that reads event logs and tickets. Automate is the screenshare build. Govern covers approvals and run history. The product menu also lists a Caddi Recorder, Run History, Integrations and an AI agents section whose first entries include a Billing Specialist and an Accounts Payable agent. The homepage says Caddi automates work "across all your existing tools," and the menu has a forms section covering finance filings such as ADV, CRS and U4, which fits the law and finance audience.

Control sits with humans. The homepage FAQ says an agent cannot decide something on its own: code runs the work, and changing a process takes a human approval. Each system gets one permission scope, under the access the person already had, with no standing admin. Every run replays decision by decision, with the permission each one used. If something goes wrong you can pause a loop or pull its connection, and the company says it stops mid-queue, not at the end.

Each agent owns one process and moves data between the sources of truth you already own. The company says that means the same input takes the same path every time, and contrasts this with networks of agents that each keep their own memory.

The evidence

Pricing is published on the comparison page against Appian. The model is usage-based within a plan: a monthly allowance of credits and a number of active loops. Individual is $0 and free for one user with no time limit, with 200 credits and 10 loops per month. Startup is $250 per user per month. Business is $6,500 per month. Enterprise is custom. The same page says the Team plan is for companies of up to 10 employees, that Business and above include unlimited users, and that Individual and Team can be signed up online without a sales call. It gives no price for Team, so ask before budgeting for it. The Automation Anywhere comparison summarizes Caddi's pricing as "Custom; pilots start low and scale with usage."

On security, the homepage says Caddi is SOC 2 Type II compliant, and calls this the precondition for deploying any automation at a regulated firm. Its FAQ says the contents of your connected systems are never used to train models, that OpenAI and Anthropic calls run under zero-retention terms, and that accounts under a Data Processing Addendum exclude their content from model development entirely. Those are company claims; no hosting region is stated on the pages shown.

The company describes the speed claims in specific terms: about 45 minutes to a first agent, and agents that run "hundreds of times a week" once built. A byline on the Automation Anywhere comparison credits Jason Alafgani and dates it Jun 1, 2026, updated Sep 18, 2026. No customer names, user counts or integration counts appear on the pages shown, so treat the "across all your existing tools" line as a claim without a list behind it.

Who it suits and who it does not

Caddi suits an operations or admin team in a law firm or financial practice that has a repeatable job, such as filings, billing or accounts payable, and wants it running without building an app or hiring RPA developers. The Individual plan allows a single person to try it with no card and no sales call. Teams that value audit trails will like the replayable runs and the human approval step on process changes.

It is less suited to someone who wants to drag together their own logic, tune every step or own the build. The company is explicit that it builds and maintains the automation for you, which also means you rely on the vendor for changes. A larger team should expect a quote past the Business plan, since Enterprise pricing is custom, and the Startup plan's per-user price adds up as headcount grows. Finally, the pages describe a platform for ops work in regulated services, so a founder looking for a website, store or customer-facing app builder is in the wrong place.