Personal AI Kukku Me Platform Guide

Platform Model

• Available on iOS.
• Agent-first architecture for specialized workflows.
• On-device-first processing with optional external tools and models.
• Modular capabilities: tools, skills, documents, nearby collaboration, profile and backup.
• Home Dashboard surfaces a Smart summary card, Reflection card, CTA agents and horizontal rails; an All Agents view supports grid and card layouts.

Skills, Journals & Analytics

• Skills are capability modules following the skills.md standard. Kukku ships two native skills — Habit Tracker and Expense Tracker — and additional skills can be added.
• Skills and tools are mutually exclusive on a single agent; an agent is either skill-based or tool-based.
• Journals are a built-in feature within relevant agents, presenting data in today, weekly and monthly views. Analytics runs internally. Neither is a skill.
• The Smart card produces a plain-English daily summary; the Reflection card provides an evening check-in.
• Skill orchestration runs in Analyzer mode. See the Skills Guide.

Routing and Execution

• Analyzer mode required for tool orchestration and planned multi-tool workflows.
• Direct mode provides low-latency direct responses only; tool flows are disabled.
• Optional deterministic workflow sequencing per agent.
• Keyword-guided tool selection with rule-based safeguards.

Premium Subscription: Required for Analyzer mode and all tool orchestration.

Model Setup and Memory Planning

• Local AI is powered by community models optimized for Apple MLX.
• Higher-memory configurations can expose a Gemma 4 E2B model option (effective 2B parameters, ~5.3B actual per Google) for stronger on-device results.
• On lower-memory devices, use the default low-memory model profile.
• Selecting oversized models on low-memory devices can cause unpredictable responses and app instability, including possible crashes due to memory pressure.

Output synthesis controls: Advanced settings (temperature, top-p, repetition controls, output token limits) directly affect response style and length. For practical tuning steps, see User Guide and Tools Guide.

Recommendation: Keep the default low model selection on lower-memory devices unless you have validated stability for your workload.

Remote and External Models

• Escalation mode can route a task to a more powerful model when needed; Kukku prompts for consent before sending data off-device.
• Agents can connect to a remote model or an OpenAI-compatible external endpoint configured with the user's own API key, base URL and model name.
• Only the prompt and context for the request are sent to the configured provider; local-first remains the default path.

Data note: External inference and search rely on user-supplied credentials; the user is the data controller for those providers.

Experimental: MCP Client Tool

• Early support for the Model Context Protocol (MCP) lets an agent call tools exposed by a configured MCP server over an HTTP transport.
• Enablement: off by default. Turn on Model Setup → Tooling Features → Enable MCP Client Tool to make the MCP Client tool type available in Tools.
• Connection: configured per tool with Server URL and an authentication mode — Bearer token (recommended), Header API key, Query API key, or None — plus API key/token and optional secret.
• Governance: Advanced options include extra headers (JSON), include/exclude tool filters, read-only-only, a max-tools-exposed cap, and a discovery mode (Simple / Progressive / Aggressive).
• Credentials are stored on-device and redacted from backups. Experimental capability; behavior and configuration may change across releases. Full walkthrough in the User Guide.

Tool Enablement

• Tools are attached per agent, not globally assumed.
• Credential-backed tools require valid provider setup.
• Tool packs can be imported to expand agent capability sets.
• Operational checks should include permissions + credentials + assignment.

Knowledge Workflows

• Supported file types: PDF, images, text.
• Document processing must complete before reliable retrieval.
• Comparison queries should reference both source documents explicitly.
• Grounding an agent with an advisory document biases its responses toward that guidance — see How to Influence Agent Output for sample advisories to download and upload.

Nearby Agents

• Peer discovery, secure pairing, and remote agent bridging to dashboard.
• Password-based secure connection flow.
• Designed for local collaboration and handoff workflows.

Premium Subscription: Required for Near By Kukku Agents.

Nearby Bots (OpenClaw Compatible)

• Local gateway discovery and approval flow via gateway token.
• Remote bot-agent listing and add-to-dashboard bridging.
• No special Wi-Fi SSID naming requirement.

Premium Subscription: Required for Near By Kukku Bots. Operational requirement: bot gateway must be discoverable on local network and approved in the app flow before agent listing.

Quick Agent Setup

• Quick Agent Setup accelerates creation with guided defaults and tool recommendations.
• Agent editing supports role/personality refinement and tool/routing updates.
• Designed for broad device compatibility.

Team Agents

• Team Agents support multi-agent orchestration for advanced workflows.
• Recommended on high-memory devices for best performance consistency.

Output Experience

• Kukku can render specialized result cards (for example health and stock views).
• Structured output is surfaced in chat alongside natural language responses.
• Raw protocol traces remain hidden in normal user mode.

Data and Safety

• Core design prioritizes local processing and user-controlled integrations.
• Backup/restore supports continuity while protecting secrets.
• API credentials are redacted from portable backups.