Project Doggo

Simple things should be simple, complex things should be possible.

Alan Kay

TL;DR

Doggo Chat makes a Kagi expert available throughout Kagi. Doggo automatically resolves member problems when possible or prepares feedback developers can act on.

Kagi Mods alter an individual's Kagi without changing the shared product. Doggo helps author Kagi Mods, but members can also author Mods directly.

For unusually valuable personal work beyond self-service, members can buy or gift a bounded block of time from a human Kagi expert.

Members get better help without waiting, while developers regain time and attention. Kagi becomes a platform for personal software and gives members a stronger reason to subscribe.

Today's Kagi

Members often know the outcome they want, but not whether Kagi can already deliver it, which part of Kagi to use, or what Kagi should build if it cannot. If developers have to guess, they risk solving the wrong problem. Consider three members:

• ZeroResults couldn't figure out why their query +1(405) location got no results. Was this a Kagi bug?
• OffCenter waited five days to learn that Kagi already supported their setting.
• NicheSidebar requested a search-settings sidebar. Kagi deemed it too niche to justify the added clutter, but Custom CSS could fulfill it.

Kagi Assistant might help, but members must know to ask Assistant, explain the problem, decide whether its advice fits Kagi, and carry out the fix themselves.

Kagi takes member feedback unusually seriously, but every request consumes developer attention. Someone must understand the goal, investigate it, and choose the right path, even when the answer is simply an existing feature or justified Nofix.

When members give up, accept a workaround, or solve a problem for themselves without reporting it, Kagi cannot see whether other members share the same need.

Kagi asks members to pay for better versions of familiar services, so "free and good enough" remains a powerful alternative. But reasonable member needs can conflict: what makes Kagi indispensable to one member may be a deal-breaking distraction for another. Kagi's small team can neither add every feature to the shared product nor build a separate version for everyone.

In a Perfect World

• ZeroResults immediately learns why the + probably led to zero results and gets links to queries that return results.
• OffCenter gets the centered search results they wanted within a few seconds.
• NicheSidebar gets a search-settings sidebar implemented as Custom CSS, without knowing CSS or adding clutter to everyone else's Kagi.

Members can start with a problem or goal without a concrete feature proposal. They respond to suggested options and shape a candidate design via conversation. When Kagi itself must change, developers receive the design and the problem it is meant to solve.

Developers spend less time explaining existing features, reconstructing requests, and guessing what members want. Kagi can see which needs recur and decide what to explore next.

When self-service is not enough, members can buy a bounded block of time from a human Kagi expert for themselves or someone else. The expert works toward an ambitious personal outcome. The payment helps fund Kagi's team and shared product. Kagi can show the resulting artifact to other members and learn whether they share the problem and find the solution useful.

More value for one member no longer has to mean more complexity for everyone. Kagi can offer something beyond better versions of familiar services: a Kagi that fits each member and can grow into a platform for personal software. That gives members a reason to subscribe even when "free and good enough" is available.

How Doggo Helps

Doggo is a Kagi expert available throughout Kagi. Members can open a chat from any page, and Kagi can offer gentle hints when Doggo may help.

Doggo uses the current page, query, and relevant settings to restate the problem and desired outcome. Doggo asks clarifying questions when needed, then takes the simplest useful path:

1. Diagnose the current search (ZeroResults).
2. Surface and use an existing Kagi feature (OffCenter).
3. Build a Kagi Mod with the member (NicheSidebar), with optional paid help from a human Kagi expert when self-service is not enough.
4. Draft feedback with the member when Kagi itself must change.

For OffCenter, the match is clear and Kagi has a setting for it, so Doggo can answer right away:

The settings token is a link that opens and highlights the control. Doggo stays open whether the member follows it or navigates to Settings manually. Reopening the browser discards any session-only changes.

If the request requires a change to Kagi itself, Doggo turns the conversation into a feedback draft. It includes the desired and current behavior, what the member tried, any candidate design they developed together, and details Kagi staff would otherwise need to ask for, such as the query, settings, platform, browser, search region, screenshots, and errors. The member edits and approves the draft.

When a change would help one member but add clutter to the shared product, Doggo can build a Kagi Mod with them. A Kagi Mod changes one member's Kagi without changing the shared interface. For NicheSidebar, Doggo confirms how the page should look, writes the Custom CSS, and previews the result. The member responds; Doggo revises the CSS until it works, then applies the approved version to the member's Kagi.

When self-service is not enough, a member can buy or gift a bounded block of time from a human Kagi expert. Any resulting personal artifact remains inspectable and uses only capabilities approved for Doggo. Payment buys expert time toward a personal outcome; it does not guarantee a result or give the member control of Kagi's shared roadmap. The resulting artifact shows what the member wanted and how one solution works. Kagi can ask whether other members share the problem before deciding whether the solution should remain a Mod or become a feature Kagi builds and supports for everyone.

The Kagi Mod model unifies Custom CSS, Lenses, and Bangs with a common format, permission model, and controls. Each mod is a plain text file that defines what it does, where it runs, and what data or capabilities it needs. Like a Tampermonkey userscript or browser extension, it can be individually turned off, edited, exported, or removed. Future Kagi APIs could expand what mods can do, allowing them to add custom widgets or transform search results. A custom widget could adapt a standard Kagi widget or be built from scratch.

Members can create and manage mods directly, without Doggo. Doggo is an optional conversational interface to the same Kagi-approved tools. When Kagi adds a new Mod API, members can use it directly or ask Doggo to help.

Members can control mods without leaving the search page. A Mods tab beside Chat lists the mods active on that page, with a toggle beside each name. A dedicated Mods settings page handles the complete collection and more involved editing.

A mod can solve one small need or grow into personal software. A member could build a desktop research tool that searches Kagi, local files, notes, bookmarks, and other services together, then saves useful results for later. That gives the member more value from Kagi without adding clutter to the shared interface. That gives members a reason to subscribe even when free alternatives are good enough.

Privacy, publishing, and usage data are separate decisions. A mod can stay private with no usage data collected. Usage-data collection may be controlled by Kagi, the member, or both. A duckpower conversion or random-number widget with a lower bound could stay private or be published without adding clutter to everyone's Kagi. A member could also publish a stock graph or historical currency graph, helping other members before Kagi updates the shared product. The stock graph request alone has 57 votes and has remained Planned since 2022.

Shared mods give Kagi more than votes. Installs, sustained use, remixes, and ongoing maintenance show whether members keep finding a mod useful. Kagi can then decide whether a solution should remain an optional mod or become a feature Kagi builds and supports for everyone. One shared Kagi core can support many Personal Kagis.

Common Questions

Can current AI make useful personal software?

Personal software just has to do what the member asked; it doesn't need full polish or support for every edge case. After three years of trying note apps, Andrew Warner used Lovable to make the narrow tool he wanted in 15 minutes. A coding agent also turned the NicheSidebar request into working Custom CSS for Kagi.

Why include paid expertise?

When self-service cannot finish valuable work, a member can pay a Kagi expert to turn their need into a working solution. A small number of high-value engagements could provide meaningful revenue for Kagi and help fund the shared product. Free-to-play games demonstrate this revenue model: a small fraction of players generates most purchase revenue, subsidizing the experience for everyone else. Complexly is closer to Kagi in purpose: supporters can spend thousands on Crash Course coins that help fund education available to everyone.

Paid expertise also helps Kagi discover product opportunities. Each finished solution gives Kagi a concrete way to test whether other members share the problem. Paperless Pipeline followed this pattern: its founder validated one customer's unmet need with other brokers before building the product. Kagi can use the same evidence to decide whether a solution should remain a Mod or become a supported feature.

What is the smallest useful version of Doggo?

The MVP can begin on Kagi Feedback. Past discussions provide test cases, while new discussions give Doggo a live, staff-reviewed channel. This requires no new member-facing interface or Mod APIs. If it works, Kagi can pilot Doggo inside Kagi before investing in new APIs or the larger personal-software platform.

Won't Doggo cost too much to operate?

Doggo runs only when a member needs help, not on every query. Once Doggo improves a query, changes a setting, or creates Custom CSS, the result can keep working without more model calls. Simple tasks can use cheaper models, reserving stronger models for generation and repair. Kagi already caps Assistant usage at each plan's value and adds a 20% operating margin to token costs; Doggo could use the same system. The pilot should work at today's prices. Future price drops or local models would improve its economics but are not required.

Will Kagi have to support every Mod?

Kagi would support the shared product and Mod APIs, and provide best-effort support for personal Mods. This extends Kagi's Custom CSS model: Kagi provides an editor and a no_css escape hatch if member CSS breaks. Doggo can attempt to repair a broken Mod, while the member can disable it without affecting Kagi's core. Members preview and approve changes before they are applied.

Let's build Doggo. Start with the Kagi Feedback MVP, then bring Doggo into Kagi and expand it as each stage proves useful: from Kagi expertise to safe action to personal software shaped around each member.