I built an iPhone app called Casberi. The short version: connect your apps and everything lands in one private feed on your phone — no accounts, no server, all on device. Pin what matters, tag what's yours, and execute actions directly in the platforms they originated from. One focal button handles queries, local orchestration, and quick jumping into your workflows. Because it utilizes a fully serverless, local-first architecture, your data never leaves your device unless you explicitly dispatch it. Try it out on TestFlight.
Just connect. Zero backends and zero sign-ins — the client works immediately upon installation. You bridge your local and remote apps: Mail, Photos, Calendar, crypto wallets, GitHub, Farcaster, Bluesky, RSS feeds, and dozens more. Everything they generate aggregates into a flat, chronological feed — one screen, zero tab navigation. Filter state cleanly with top-level interactive chips. Screenshot assets undergo local OCR on-device, making visual data fully queryable. Fresh live data feeds into the experience programmatically — trending token vectors and real-time NFT mints appear without external discovery loops.
Keep tabs. Pin critical indexes onto a custom dashboard — live multi-chain wallet balances, travel schedules, or recent social casts. The dashboard prioritizes chronological deadlines first, surfacing upcoming meetings and due tasks. On launch, a single-chip recap encapsulates all inbound events that resolved while you were offline. Feeds are highly contextual: the calendar presents as a timeline agenda, photos render as an image grid, and wallets present aggregate balances across multiple addresses simultaneously. You can monitor specific token price boundaries, surfacing alerts directly back into your main flow. Natural tags group disparate items laterally, eliminating nested folders.
Take action. All functional interactions route through a persistent action button in the corner. Query your historical feed using natural local language — "What did I save this week?" — computed locally on your device. Orchestrate structural reorganization via bulk commands — "Tag these as trip," "Rename work to projects" — showing an exact dry-run diff before mutation. By bringing your own API keys, you can programmatically escalate complex evaluations to remote models (Claude, GPT, Gemini, or Venice) executed peer-to-peer from your client. Resolve items directly at the origin point: status modifications on platforms like Todoist or GitHub execute cleanly with strict opt-in confirmations. Casberi integrates deeply with iOS Shortcuts and Siri for frictionless hands-free ingestion.
Make it yours. Fully personalize the interface with custom backgrounds and layout options. Your data remains strictly sandboxed and syncs peer-to-peer only upon user activation. Import and export formats are entirely transparent, featuring multi-layered wiping options to immediately revoke integrations and purge local device databases. Integrations for web3 wallets run in strict read-only mode, assuring cryptographic keys and funds remain untouched.
Modern consumer software suffers from severe user-data and execution fragmentation. Every user interaction generates isolated, platform-specific schema artifacts trapped behind proprietary APIs and remote DBs. Navigating this ecosystem requires compounding context switches across separate client binaries. While personal AI systems promised automated coordination, they have instead materialized as disconnected SaaS chatbots, introducing another layer of isolated data sandboxes, redundant user states, and tracking endpoints.
Looking at the ecosystem, the core friction points sort into eight clear engineering and UX failures:
Copy-paste as standard integration layer. To finish anything, you copy, switch apps, and paste. And if AI agents do your work, your devices lack the local system context to coordinate them.
Scattered capture targets. The downstream storage schema depends entirely on the capture channel, partitioning your personal telemetry.
Mutilated schema indexing. Maintenance requires manual, redundant filing tasks and manual sorting across multiple proprietary tables.
High-latency search routing. Locating historical assets requires first remembering which native application holds them, introducing significant visual retrieval latency.
Fragmented execution loops. Remote LLM context cannot directly write state changes or execute mutations within local applications, stalling workflows.
Non-portable configuration states. Casual users bypass optimization entirely. Power users must reconfigure environments per machine, preventing portable setup or cross-device state preservation.
Cold-start contextual agents. Newly initialized model runtimes carry zero historical user telemetry, localized context, or cross-application permissions.
Silent telemetry failures. Background operations and remote state mutations complete or fail in silence without streaming execution telemetry back to the primary feed.
Translating these failures into a tight engineering specification requires moving past surface cosmetic adjustments. Any robust, functional solution is bounded by clear system design constraints:
Homogeneous database & local storage. All telemetry (notes, links, screenshots, events, chat logs, voice inputs) is stored on-device using a unified SQLite/CoreData schema inside a single disk index. OCR runs fully offline using local Apple Neural Engine APIs.
O(1) index-level retrieval. Omnipresent schema-wide search targets the exact database resource directly, completely bypassing nested folder paths or remote network requests.
Continuous zero-gesture capture. Passive, background ingestion through native share extensions and device-level event loops. Connected apps stream directly to the local storage layer without active human management.
Integrated ingestion-to-outcome vectors. Every ingested record exposes immediate inline execution pathways. Calendar metadata parses immediately into structured local reminders, collapsing the gap between capture and action.
Deterministic state preservation. Value is delivered immediately out-of-the-box. Sensitive access tokens and operational secrets are securely written directly to the localized hardware Keychain, preventing any setup drift.
Interface utility over agent personas. The system functions strictly as an interactive editor and compiler, not an anthropomorphic conversational bot. No synthetic avatars or conversational throat-clearing.
Native system haptics & presentation. Pure native client architecture (Swift/SwiftUI) using low-latency CoreAnimation pipelines. Zero cross-platform web wrappers or Javascript runtimes on the main view threads.
Explicit schema sandboxing. Transparent local-first access control boundaries. Native system APIs handle on-device telemetry, and remote webhooks are isolated individually. Visual dashboards state exactly what local tables map to your feed.
True transactional approvals. Action requests render the exact state mutation diff—depicting the initiating thread, target database schema, and post-mutation parameters—requiring a physical user tap to execute.
Unified asynchronous telemetry. Job completions, offline connection state drops, and direct API mutations stream as structured events in the main chronological list. System execution errors are surfaced explicitly.
Casberi is a fully native client. Your indices live in an on-device database; there is no cloud coordinator. Sync runs peer-to-peer on your own iCloud. Offline access is fully functional. There are zero trackers and zero downstream telemetries. Language processing is computed locally; external APIs are called directly from your hardware to individual model endpoints using your personal keychains.
I omitted several systems by design. There is no custom note editor — Casberi reads but leaves editing to your existing applications. There is no central task dashboard — action states remain linked directly to the objects they resolve. There are zero gamification mechanics, and crypto interactions run strictly as read-only.
Casberi is open to build feedback on TestFlight prior to its public App Store launch. If these coordination failures resonate with you, I would value your critical feedback: https://casberi.app