SYNCED RECALL · HELP & GETTING STARTED

Your guide to Synced Recall

From your first recording to choosing AI models. Available offline in the app. External provider links need internet.

Updated September 7, 2026. Provider availability, prices and terms can change.

Get started

Record a short test first. On Today, choose Manual, grant microphone permission, start recording, then stop. Open Library, select the recording and play it. A saved recording and a completed transcript are separate states. If processing fails, keep the audio and inspect Processing Center. You can use local features without creating an account.

Back to contents

What each AI task does

Transcription turns audio into text. Summarization turns that text into an overview. Ask retrieves relevant Memories and answers questions using their content. Action extraction identifies supported follow-ups. Select a provider and model separately for each task. A language model is not automatically an audio transcription model. Claude is used for text tasks by the current adapter; choose a speech provider for transcription.

Back to contents

What is an API key?

An API key lets Synced Recall use a service under your provider account. It is different from your password and may permit billable usage. Create the key on the provider’s official website, copy it, and paste it into AI connections on the phone or Settings on the paired desktop portal. Never paste keys into recordings, chat messages, shared spreadsheets or screenshots. Synced Recall stores saved keys encrypted on your phone. The portal receives key-presence information, not the saved secret.

Back to contents

Connect a provider: save, test, select

1. Open Settings → AI & transcription and choose the connection. On desktop, open Settings after pairing. 2. Follow the official key link and create a key in your account. 3. Choose the matching service before pasting its key. 4. Save. 5. Test the saved key or load models when offered. 6. Choose the task, provider and compatible model, then use/save that selection. A successful metadata test checks authentication or model access, not available credit, billing or the quality of a real transcription. Some adapters have no metadata test; they say so. Verify those with an intentional short recording.

Back to contents

Choose and refresh models

Use the provider’s live model list where available. Model availability can depend on your account, region and permissions. Refresh after saving or replacing a key. Exact selections stay pinned; the app should surface a missing/retired model instead of silently switching it. Automatic routing is a separate choice. OpenAI-compatible connections have a shared credential slot, so replacing OpenAI with Groq or xAI in that slot replaces its previous key. Anthropic, Gemini and OpenRouter have separate slots. Not every model a provider advertises is supported by this app’s current API adapter.

Back to contents

Free tiers, credits and paid usage

A consumer ChatGPT or Claude subscription is separate from API usage. Free tiers have quotas; promotional credits can expire; a free website is not proof of a free API. Free-only routing is an app policy, not a guarantee that your provider account cannot be charged. Review billing in the provider console, keep account-level spending limits where available, and set Synced Recall’s routing budgets. Estimated costs are not a substitute for provider billing. A 429 response can mean rate limits or exhausted quota; repeated retries do not create new credits.

Back to contents

Cloud processing and privacy

Offline Only blocks cloud processing. Other presets may use your configured cloud providers according to routing choices. Saving a key can make it available to automatic routing even if you have not pinned it. Cloud transcription sends audio to the chosen speech service; cloud summaries and Ask send the relevant text/context to the chosen model service. Review each provider’s current data terms. In particular, Google AI Studio unpaid-service terms differ from paid-service terms. Do not equate a free tier with private processing.

Back to contents

Manual, Smart, Continuous and Scheduled

Manual capture starts and stops when you choose. Smart/Continuous use the app’s capture pipeline and speech detection; the microphone may remain active while listening for speech even when no Memory is being saved. Short pauses can finish a captured span. Scheduled capture follows its configured window subject to Android restrictions. The notification and Today status show the actual state. Android may require a tap after reboot or force-stop. Check permissions, battery restrictions and Processing Center when something stops unexpectedly.

Back to contents

Small slices and longer recordings

Library groups recordings into expandable 15-minute time groups by default. Choose 30-minute groups or Individual on phone or desktop; each surface remembers its own view. Grouping does not merge audio or create a combined summary. Search and selection can expose individual records. Transcription request sizes are separate: remote processing uses at most ten minutes per request, or a smaller adapter limit. Local engines keep their own limits. Requests use overlapping chunks for long files. This budget is not a promise of optimal speed or accuracy for every model.

Back to contents

Organize, play and find

Use folders to create, rename and move recordings on phone or desktop. Deleting a folder keeps its recordings in the library. Favourites preserve your marked choices. Open a Memory for its audio, transcript, summary and supported actions. The desktop waveform comes from actual audio samples. Play/pause, skip and seek control the selected recording. Use Ask with a specific question, date or context; inspect the linked source Memories rather than treating a generated answer as independent evidence.

Back to contents

Pair a computer

Open https://app.syncedrecall.com on the computer. On the phone, enable Settings → Connections → Desktop Access, then scan the QR shown on the computer and approve access. If the camera cannot scan, expand Camera not working? on the computer and paste the complete private link into the phone. The fragment after # is part of that short-lived private link; do not share it. Choose Remember or 7 days for reconnection. Explicit revocation on the phone removes approval. Both devices still need reachable same-network connectivity; remembered trust does not provide a remote relay.

Back to contents

Desktop connection troubleshooting

Keep Desktop Access on and the phone available. Reconnection may take about a minute after network restoration. Check Wi-Fi, VPN, guest-network isolation and firewall settings. A revoked or expired approval needs new pairing. If you clear the browser’s site data, its remembered trust is removed. The desktop service’s long-duration Android limits are still under development; do not treat a successful short session as an all-day availability guarantee.

Back to contents

Export, archive and Google Drive

Current Memory export and archive features are different from full settings/new-phone backup. Choose and verify a destination before relying on it. Check archive status and errors; never assume a queued upload is complete. Google Drive requires account authorization on the phone. Changing phones requires a portable restore format; copying local Keystore-encrypted files alone is insufficient. Full automatic account backup, selectable backup levels, new-phone restore and custom account creation remain planned, not available controls.

Back to contents

Updates, credentials and recovery

Ordinary in-place app updates preserve settings and credentials. Do not uninstall or clear app storage to update. Database migrations must preserve data; unsupported migrations fail rather than silently wipe it. Uninstalling, device loss or losing local encryption keys is different from updating. Planned recovery offers Standard Google-account backup and optional Private recovery-key protection, but this has not shipped. An optional app PIN is also distinct from a portable backup recovery key.

Back to contents

Processing errors and slow jobs

Open Processing Center for the actual job state and error. Check that the task has a supported provider/model, the key is saved, required local models are downloaded, and Wi-Fi/charging constraints are satisfied. Authentication failures need a valid key; quota failures need account quota or another explicitly selected provider. Large audio, device temperature, model size and network conditions affect duration. Retry only after addressing the cause. Failed transcription should retain audio; failed summary should retain the existing transcript. Avoid deleting the source while investigating.

Back to contents

Available versus planned

This guide describes current controls and explicitly identifies gaps. Monthly Google Sheets journals, monthly topic reports, automatic portable backups, Google/new custom account onboarding and cross-network desktop relaying are not yet completed. Group-level combined summaries and play-all are also pending. No setup instruction should require a control that is merely planned. Report the screen, action, visible error and whether source audio is still available when describing a problem; exclude API keys and private pairing links.

Back to contents

Official provider links

Use official consoles only. Start with a short test and inspect current pricing before sending longer recordings.

OpenAI pricing · Claude pricing · Gemini pricing and data use · OpenRouter limits