Help
Answers to the questions people ask about Sparks.
How does GitHub sync work?
Enter your GitHub username in Settings. Every public repo becomes a project, with its description, language and live site if the repo has one.
For private repos, or to avoid GitHub's rate limits, add an access token. Create a fine-grained token with read-only Metadata access to all repositories. The token is stored encrypted and is never shown again. You can replace or remove it at any time.
With “Keep in sync automatically” on, new repos appear within a day by themselves, and Sparks also checks when you open Notes or Projects, at most every six hours. You can press Sync now whenever you like. Forks are left out by default, and archived repos are included. Both are switches in Settings.
Syncing only reads repo details. It never changes your repos, and your notes and your own edits to a project are never overwritten. If a repo is deleted on GitHub, its project stays, with its notes.
Can a message thread become a to-do list?
Importing a chat creates notes from its messages. You can add subtasks, due dates and urgency, and mark notes or subtasks done.
Sort with AI suggests projects and titles during import. It does not extract action items from a whole conversation or find semantic connections between notes. You can ask a connected agent to read a thread and create a task list.
Save a new idea with ⌘ Enter on Mac or Ctrl Enter on Windows. Existing notes save automatically. Search and filters stay in the address so Back returns to the same results.
How do I tell my agent not to build an idea yet?
Add the tag #hold (or #dont-build) to the note. Connected agents still see the note, but they skip it when looking for work until you remove the tag. Build is refused while the tag is on. Plan and Research can still run once the agent has enabled them.
How do I export chats from my messaging apps?
Export the chat where you message yourself, then pick the file on the Import page. You choose which messages to keep, and only those are saved. Menu names change now and then, so the steps below may differ slightly on your device.
- Open the chat with yourself (or any chat).
- iPhone: tap the chat name at the top, then Export Chat. Android: tap the three dots, then More, then Export chat.
- Choose Without media.
- Save the file to Files or Drive. You get a .txt file, or a .zip with one inside.
Telegram
Chat export is only in Telegram Desktop on a computer, not in the phone app.
- Open the chat, for example Saved Messages.
- Click the three dots at the top right, then Export chat history.
- Untick photos, videos and other media.
- Set the format to Machine-readable JSON, then click Export. You get a file called result.json.
Messenger
- On facebook.com, open your profile menu, then Settings & privacy, then Settings.
- Open Accounts Center, then Your information and permissions, then Export your information.
- Create an export, choose your Facebook profile, and pick Customize information.
- Tick only Messages. Set the date range, set the format to JSON, and start the export.
Meta prepares the file and tells you when it is ready to download. That can take from a few minutes to a few days.
- In the Instagram app, open your profile, tap the menu, then Accounts Center.
- Open Your information and permissions, then Export your information.
- Create an export, choose your Instagram profile, and pick Customize information.
- Tick only Messages. Set the format to JSON, and start the export.
Instagram emails you when the file is ready.
Recall, your 6x7 chat importer
If you use Recall (recall.6x7.gr), skip the files. Add its address in Settings and Sparks pulls the messages you sent yourself.
Record an idea (voice)
Open the Record page and tap the microphone. On Android you can long-press the installed app icon and choose Record idea (it opens /record?auto=1).
iOS Shortcut (Back Tap or Action Button)
- In Settings → Voice capture, tap Create shortcut token and copy the token. Paste it somewhere safe (Notes) for the next step.
- Open the Shortcuts app. Create a new shortcut named "Record idea".
- Add Record Audio. Set "Stop Recording" to On Tap (or a short duration you prefer).
- Add Get Contents of URL. Method:
POST. URL:https://sparks.6x7.gr/api/ideas/voice. Request body:File(the audio from the previous step). Add a headerAuthorizationwith valueBearer YOUR_TOKEN(paste your token). - Optional: add Show Result so you see the saved note id, or Open URL to
https://sparks.6x7.gr/noteswhen finished. - In Settings → Accessibility → Touch → Back Tap (or the Action Button), assign this shortcut. Triple back tap now records and sends the idea without opening Sparks.
The shortcut needs network access. Recordings are transcribed on the server, tidied with AI when available, and filed under the best matching project (you can change the project afterwards in the app).
What does Analyze idea do?
On a note, Analyze idea reads what you wrote and suggests a short summary, a few questions, and steps. It sends that note to Anthropic's Claude API. Nothing is saved into the note until you tap Add these steps. The same text is not sent again: reopening the note shows the saved read. You pick the Claude model next to the button, or in Settings → AI. Changing the model does not spend a call until you read again.
Plan, Build, and Research appear next to the note only after your agent says it understands them. Until then the control stays Send to agent. Build always asks you to confirm, and it is refused while the note is tagged #hold.
What does Sort with AI do?
On the Import page, Sort with AI suggests a project and a short title for each message you selected. To do that it sends only the selected messages and your project names to Anthropic's Claude API. The model is the one chosen under Settings → AI for sort and voice.
It runs only when you press the button. You can change every suggestion, and nothing is saved until you import.
For developers: the interop API
Other apps from 6x7.gr can read your projects and send ideas into Sparks. All of them use the same sign-in, so one access token works everywhere. The full spec is the INTEROP.md file in the repo; this is the short version.
Authentication
Send Authorization: Bearer <access token> on every call. The token is the Supabase access token of the signed-in user, from the shared 6x7 project. The app verifies it and runs every query as that user. There are no cookies on these endpoints. Browser calls are allowed from https://*.6x7.gr.
Endpoints
GET /api/interop/v1/manifest: public discovery document with the app name, version and endpoints. Open it.GET /api/interop/v1/projects: the user's projects, most recently active first. Optionalinclude_hidden=1andstatus=active,idea. Each item hasid,name,description,status,url(where the user wants a click to go),live_url,repo_urlandnote_count.POST /api/interop/v1/notes: send up to 500 ideas or messages. Safe to repeat: items are deduplicated onsource_appplusexternal_id. Each item needsexternal_idandtext, and may addproject_id,project_hint,tags,created_at,platformandurl. Aproject_idwins, thenproject_hint, then a guess from the text. Otherwise the note lands in Unsorted (notes with no project). The answer is{ "inserted", "skipped", "rejected" }.
Sparks also reads GET /api/interop/v1/messages from your chats app, with since, self_only, limit and cursor. It returns items with a stable id, text and sent_at, plus next_cursor.
curl https://sparks.6x7.gr/api/interop/v1/projects \
-H "Authorization: Bearer $ACCESS_TOKEN"
curl -X POST https://sparks.6x7.gr/api/interop/v1/notes \
-H "Authorization: Bearer $ACCESS_TOKEN" \
-H "Content-Type: application/json" \
-d '{"source_app":"chats","items":[{"external_id":"wa:3EB0C4","text":"catchy: add speaker colours to the export"}]}'Servers calling these endpoints should refuse redirects and only call https://<name>.6x7.gr hosts, so the token never leaves the family of apps.
Still stuck? Write to help@6x7.gr.