@mushi-mushi/core

Your AI wrote it. Mushi tells you why it broke.

The shared types, API client and utilities under every Mushi Mushi SDK.

You probably don't need to install this directly. Run npx mushi-mushi and the wizard installs the SDK for your framework (react, vue, svelte, angular, react-native, capacitor or web), which depends on this one. Reach for core when you build your own adapter or need the types.

flowchart LR
    A["Framework SDK<br/>react, vue, web…"] --> B["@mushi-mushi/core<br/>types · API client · queue"] --> C["Mushi API"]

What's inside

| Piece | What it does | | --- | --- | | Types | MushiConfig, MushiReport, MushiEnvironment and every shared interface | | API client (createApiClient) | Fetch with retry and backoff. Tags its own requests with X-Mushi-Internal so SDKs never report on Mushi | | Pre-filter (createPreFilter) | On-device spam and gibberish filter, before anything is sent | | Offline queue (createOfflineQueue) | IndexedDB queue that syncs on reconnect | | Environment (captureEnvironment) | Viewport, user agent and client hints, connection, accessibility preferences, page-load timing | | Identity | Anonymous reporter token (getReporterToken) and tab-scoped session id (getSessionId) | | Rate limiter (createRateLimiter) | Token bucket that stops an SDK flooding the API | | Breadcrumbs (createBreadcrumbBuffer) | 50-entry ring of recent events, in Sentry's breadcrumb shape | | Errors (normaliseThrown) | Turns any thrown value into { name, message, stack?, cause? } |

Presets

Pass preset instead of setting every widget, capture and proactive flag. Your explicit config always wins, key by key.

| | minimal | standard | full | | --- | --- | --- | --- | | Console capture | on | SDK default | on | | Network and performance | off | SDK default | on | | Screenshot | on report | SDK default | automatic | | Session replay | off | SDK default | lite | | Proactive prompts | none | SDK default | all on |

The postures production-calm, beta-loud, internal-debug and manual-only also work. Precedence, highest first: explicit config → preset → env vars (NEXT_PUBLIC_MUSHI_*, VITE_MUSHI_*) → SDK defaults.

import { expandPreset, validateConfig } from '@mushi-mushi/core';

validateConfig(config);                // warns on typos or a bad preset, never throws (MUSHI_SILENT=1 hides it)
const resolved = expandPreset(config); // preset → nested options, explicit values win

Sampling and filters

Sampling applies to automatic error reports, and the error filters to automatic reports and captureException(). A report a user sends from the widget is never sampled or filtered out. Only your own beforeSend can change or drop it.

| Field | Default | What it does | | --- | --- | --- | | sampleRate | 1 | Share of automatic error reports to send (0–1) | | replaySampleRate | 1 | Share of sessions that record replay | | ignoreErrors / denyUrls / allowUrls | — | Drop matching errors from automatic capture and captureException(). URLs match the frame that threw | | beforeSend | — | (report) => report \| null for every report, after the PII scrubber. Return null to drop | | tunnel | — | Same-origin ingest path (e.g. /api/mushi-tunnel) so ad blockers don't drop reports |

Full option reference: Core SDK docs · Presets · Runtime config.

License

MIT