I've been using Inkdrop for the last 2 years. Kuku gives me a much better experience when editing markdown files. Love the AI integration and horizontal tabs at the top when working within multiple files. First experience is great so far.
Would love to get some reassurance on security and privacy as I add a lot of private information into my notes apps.
@mansuiki Instead of (or additionally to) bring your own key, it would be cool if you had an option for using ACP instead so it can use claude-code/codex headlessly under the hood so we can use our plans. It also lets use the agent's tool use features (for on the fly math with python scripts or whatever else we wanna do with our notes)
@bigmacfive Congrats on the launch! Does it support custom Gemini API keys or is it bundled into the app price?
@mohsinproduct Thanks! Custom Gemini API keys aren’t supported yet, but it’s coming very soon 👍
@mohsinproduct We're also planning "Bring Your Own AI/API" support to allow other model providers in the future
@bigmacfive
The “Obsidian + Cursor, but native” framing is perfect. The diffs idea is a big deal too - once an agent can edit files, reviewability becomes the whole product.
How are you thinking about guardrails for the agent - is it scoped to a vault/folder by default, and do you plan to add a “dry run” mode where it proposes changes before touching files?
@dmitry_petrakov Exactly — reviewability is the product. The agent is scoped to the current vault/folder, and edits always go through a diff-based review. A true “dry run” / propose-first mode is very much aligned with our thinking and on the roadmap.
@bigmacfive This is the right approach: scope it, then make every change reviewable. Propose-first mode is the step that makes it feel safe. How are you imagining the review UX - a list of suggested diffs you can cherry-pick, or a single “plan” you approve?
@bigmacfive congrats on the launch. Is it BYOK or AI is included in the price?
@curiouskitty Sync is often the "final boss" of local-first apps.
Here is our current stance:
Sync Strategy: We intentionally delegate sync to the tools you already trust (iCloud, Dropbox, or Git). Since Kuku is just a window onto your local .md files, we don't lock you into a proprietary cloud.
Conflict Handling: We rely on the file system. If a sync conflict occurs (e.g., iCloud creates a duplicate file), it simply appears in your file list for you to resolve manually. We don't try to hide it behind a proprietary database.
@curiouskitty We stay file-system-native for now — use iCloud/Dropbox/Git on plain markdown. Conflicts are handled by the sync layer. Whether to build first-party sync is an open question, guided by staying truly local-first.
Looks great, downloading now.
One thought. After I sign up, I cant find a download link. It only shows for signed out users. Definitely some friction here that can be fixed.
@dennisivy11 Good catch! I confirmed the issue and just deployed a fix immediately. It should be visible now. Thanks for letting me know!
congrats on the launch! are you going to be supporting plugins eventually or is kuku focused on core features only?
@mustassim Thanks! A plugin API is definitely on the roadmap, but right now we are focused on nailing the core experience and performance.
Is there a specific plugin or workflow you are looking for? I'd love to hear what is missing from your current setup.
@mansuiki I am an Obsidian user, and I actively use Kanban board, Excalidraw, Periodic Notes (for my daily and weekly notes) and Dataview (a plugin that lets me query notes like a database)
@mustassim Thanks for the detailed list! That's the "Holy Grail" of Obsidian setups.
Periodic Notes is definitely something we want to bake into the core experience.
For heavy UI tools like Kanban and Excalidraw, we'll need to get the Plugin API ready for you.
We'll keep these in mind as we build out the roadmap!
@mansuiki wish you guys all the best!
@mustassim Thanks! Plugin support is on the roadmap. We’re still exploring what makes sense within the Tauri ecosystem, but it’s definitely something we’re actively thinking about.
Love this direction - “native + local-first + plain .md” is such a strong combo.
I’m building in the same offline-first mindset (Chrome extension land), and the trust you get from “your files are your files” is hard to beat. The reversible diffs for AI edits is especially reassuring.
Curious: what’s been the hardest part about doing Obsidian-like linking/graph while keeping everything fast and truly native on macOS?
@dmitry_petrakov The toughest challenge was definitely constructing the graph and then bridging the User and AI using locally defined tools.
Getting the AI to understand the graph context and reliably trigger local tools (instead of just hallucinating actions) was the real engineering hurdle!
@mansuiki Love this answer - the graph is the hard-won context, and making the agent actually use it (not just sound like it) is the real work. Curious what ended up being your biggest win there: tool contracts, caching/indexing, or the UX around review/diffs?
@dmitry_petrakov Tool Contracts.
Retrieval gives the AI a map, but Tools give it agency. The real breakthrough was defining strict interfaces that forced the LLM to stop being a "creative writer" and start behaving like a deterministic operator.
That constraint is exactly what keeps it from hallucinating random file paths and makes it safe to run against your actual local disk.
Congrats on the launch! The local-first approach with plain .md files and reviewable AI edits is a strong contrast to cloud-heavy tools. How do you think about conflict handling and versioning when the AI is making direct file edits, especially for users who organize large or highly interlinked knowledge bases?
congrats on the launch! What type of search can it do? Does it have semantic search via embedding?
@jasonge27 Currently, we use SQLite for fast, local full-text search.
However, we are planning to support semantic search via lightweight local embeddings in the future to keep everything offline!
@jasonge27 Thanks! Right now Kuku has fast full-text search (SQLite FTS5 + BM25). Semantic search via embeddings isn’t built in yet, but it’s something we’re actively exploring.
kuku
Thanks! Inkdrop is a fantastic app, so that comparison means a lot.
Regarding privacy:
Local-First: Your files stay 100% on your disk. We don't store them in any cloud.
Roadmap: We are preparing to add "Bring Your Own Key" and Other LLM support. This will soon allow you to have complete control over the AI data flow.
Verification: We plan to open-source the code. That way, you won't have to take my word for it—you'll be able to audit the codebase yourself.