agentickit

Add an activity, meal, transit segment, or note to a specific day of the itinerary. Mutating; framework will pop a confirm modal before adding.

Manage the trip's packing list: add items, mark packed, suggest items for given conditions. The Packing page exposes the list; the agent can see and edit it via the registered packing actions.

Edit long-term user preferences: home airport, preferred currency, default trip duration, etc. Persisted across trips. Mutating; framework fires the confirm modal before any change.

Surface flight options as a picker the user chooses from. Mounts an inline UI in the chat (renderAndWait) and pauses until the user picks one or skips. Used during the Itinerary page; a flights specialist agent also exists at /agents.

Surface hotel options as a picker the user chooses from. Mounts an inline UI in the chat (renderAndWait) and pauses until the user picks one or skips. Used during the Itinerary and Trip Detail pages.

Convert an amount from one currency to another using a static rates table. Use for quick orientation only; not for live FX.

Look up cities the app knows about and pull their catalog data. The catalog is small and curated; if a destination is missing, fall back to the web-search skill rather than inventing.

Read-only SQL access to a small seeded catalog of travel gear: luggage, electronics, comfort, toiletries, apparel. Three tables (products, categories, reviews). The validator rejects anything other than SELECT. Use this for any user question about gear (prices, ratings, stock, recommendations).

Always-on style guide for the travel concierge. Defines tone, response shape, date format, currency conventions, and the multi-step rule. Other skills assume these defaults; only override explicitly when a skill specifies different behavior.

Look up current conditions or a forecast for a specific city and date. Backed by an OpenWeather proxy when OPENWEATHER_API_KEY is set, otherwise a deterministic mock so the demo works fully offline.

Search the web across four backends (DuckDuckGo, Tavily, Firecrawl, Serper). Use only for things outside the app's local data — news, current prices, opening hours, recent events, places not in the destination catalog. The local catalog skills come first.

Spawn or dismiss an inline chart panel. The panel is hidden by default and renders only when the agent calls `show_chart`. Once visible, re-calling `show_chart` updates its type and source. `hide_chart` removes it. The panel mounts inline in the chat thread (not as a separate dialog) so it stays in the conversation context.

Search the web for current information across four backends (DuckDuckGo, Tavily, Firecrawl, Serper). Each backend is its own tool. Use this skill whenever you need information you don't already have, anything time- sensitive (today's news, current prices, recent product reviews), or any fact you'd want to double-check before stating.

Attach a react-hook-form instance to the copilot via usePilotForm. The hook registers three scoped tools (set_<name>_field, submit_<name>, reset_<name>) so the assistant can progressively fill and submit the form. Use when a consumer wants AI-assisted form completion.

Install agentickit into an existing React app (Next.js App Router is the happy path; Bun / Cloudflare Workers / Hono also work). Wire the server route, the provider, and the sidebar. Use when a consumer says "add agentickit to my app" or "getting started".

Pick a model provider for a consumer's app. Covers auto-detection, the six allow-listed prefixes (openai, anthropic, groq, openrouter, google, mistral), the Vercel AI Gateway fallback, and the escape hatch for non-registered providers (Ollama, Azure, Bedrock). Use when the question is "which model string / which env var / which adapter".

Extend agentickit as a contributor. Add a hook, a component, a provider prefix, a new protocol field, or an example. Covers the expected shape of a patch (src + test + docs), the size budget (~5 kLoC; below this, shipping is cheap — growing much past it costs the "readable in an afternoon" promise), and how to validate a change before sending a PR.

Theme, relabel, reposition, and customize `<PilotSidebar>`. Covers CSS variables (the sidebar has no Tailwind, no design system), labels for i18n, suggestion chips, position and width. Also covers when to drop the sidebar entirely and build a custom UI from PilotChatContext.

Run the workspace locally, the test suite, and the todo example. Covers the pnpm workspace layout, the `agentickit` build pipeline (tsup), the Vitest config, and the most common "why is this not working" symptoms. Use when something's broken or a contributor is trying to reproduce a bug.

Register a typed, AI-callable tool via usePilotAction. The handler runs in the browser against the consumer's React state, auth'd fetch, or any other client-side capability. Zod schema is inferred into the handler's argument type. Use when adding a new capability (create X, update Y, trigger Z) and includes `mutating:` flag guidance.

Expose a slice of React state to the AI with usePilotState. The value is serialized into every model turn; when a setter is provided, an `update_<name>` tool is auto-registered so the AI can propose whole-value replacements. Use when a consumer says "show the AI my data" or "let the AI read X".

Configure `createPilotHandler` for a consumer's server. Covers the full options surface (`model`, `system`, `maxSteps`, `getProviderOptions`), how per-request model overrides work, the provider-flexibility patch, and the security envelope the handler returns on error.

Author a SKILL.md inside a consumer app's own `.pilot/` folder. Covers the scaffold CLI (`agentickit init` / `agentickit add-skill`), the frontmatter shape (Anthropic + gbrain superset), the RESOLVER.md routing table, and the binding contract that ties each SKILL.md to a registered `usePilotAction`. Use when a consumer wants to expose capabilities as editable markdown.