dassi

dassi is an AI browser agent for automating work inside the web apps a person already uses. It runs in the side panel of Chrome, Edge, or Brave and acts through the accounts that are already signed in. A user can describe a task in plain English or walk the agent through it once, then save successful work as a repeatable workflow. The product is aimed at jobs that come back every day or week, such as checking records across two systems, collecting numbers from dashboards, updating an applicant tracking system, or filling a report. It sits between a one-time chat assistant and conventional browser automation, with an emphasis on teaching a recurring job in context.

The basic flow has three parts. First, the user gives dassi a task or demonstrates how it should be done. Next, the user chooses the sites involved, supplies changing inputs, sets a schedule, and marks actions that require approval. The agent can then run on demand or when the schedule fires. It reports completion and surfaces exceptions that need human judgment instead of treating every uncertain choice as permission to continue. Saved workflows retain a plain-language plan, their input sites, approval rules, and a run history. The schedule can cover daily, selected weekday, every-few-days, or monthly patterns. If the process changes or the agent makes a poor choice, the user can correct it, edit the stored memory, pause the workflow, or walk through the job again.

dassi works through the existing browser session rather than asking for separate connectors to every service. That makes it useful for internal tools and niche web applications that don't expose an integration, as long as a person can perform the task in a tab. The site gives examples that span reconciling marketplace payouts against store orders, screening applicants against hiring criteria, writing notes back to an applicant tracking system, and gathering weekly figures from dashboards into a spreadsheet. It can handle common browser-accessible documents including Excel, Word, Google Sheets, Google Docs, and PDFs. Scheduled work happens on the user's own machine, so the browser must be open when the run is due. If it was closed, the workflow runs after the browser opens again.

The agent supports several model providers rather than tying every workflow to one model. The site names Claude, GPT, Gemini, Grok, DeepSeek, Kimi, OpenRouter models, and local options. Users can bring their own provider keys or buy plans with managed model credits. dassi also tries to reduce the cost of repeat runs. It caches context within a run and writes a small JavaScript library for sites it has learned, allowing later steps to call known functions instead of repeatedly interpreting the same interface. The workflow can be triggered from the browser, and remote interactions can notify the user through Telegram or email when a decision is required. Users can approve a whole session when appropriate instead of confirming every click one at a time.

Privacy and control are important parts of its design, although the details depend on how it is configured. Browser actions run on the user's machine, and configured provider keys plus local conversations are stored in the browser. The selected AI provider still processes the instructions and relevant page content it needs to choose actions. Managed credit requests pass through dassi's relay. Optional remote MCP access also stores task instructions and results on dassi's servers so an authorized AI application can retrieve them. Those remote task records don't expire automatically, and disabling access doesn't delete earlier records. The approval system lets a user decide which actions must stop for confirmation, but any browser agent working in signed-in accounts still calls for careful limits and review.

dassi is a good fit for operations, recruiting, commerce, and administrative teams with repetitive browser work that crosses web apps. Its differentiator is persistence. A general assistant completes the task in front of it, while a dassi workflow keeps the schedule, plan, approvals, corrections, site knowledge, and history for the next run. Access is paid after a seven-day allocation of trial credits, with no card required for the trial. Plans with included AI credits start at $25 per month when billed monthly. A lower-cost Developer plan is available for users who bring their own model keys. Pricing rises with usage and support level, so the practical value depends on how much recurring browser work a workflow can reliably replace.