Ponder

Appwrite Functions

Function configuration from appwrite.json, live Appwrite settings, and behavior derived from the function source code.

VCS deployment notes

• Each function must keep its Appwrite VCS root directory pointed at its own folder: functions/news_refresh, functions/fact_check, functions/chat_write, or functions/like_sync.
• When the root directory is lost, Appwrite builds from the repository root and npm install fails because there is no root package.json.
• Appwrite function updates can reset omitted fields; preserve runtime, entrypoint, commands, execute permissions, scopes, schedule, and providerRootDirectory together when changing live settings.

news_refresh

Purpose: fetches external news articles, normalizes them into content rows, removes duplicates, applies quality filters, and optionally prunes old content.

Field	Value
Runtime	node-22
Entrypoint	functions/news_refresh/src/main.js
Timeout	300 seconds
Execute	none; schedule/runtime only
Schedule	0 */6 * * *
Install command	npm install
Build/runtime specification	s-0.5vcpu-512mb

Scopes:

• databases.read
• databases.write
• tables.read
• tables.write
• documents.read
• documents.write
• rows.read
• rows.write

Expected input: no request body is required by the implementation. Required environment variables are NEWS_API_KEY or NEWSAPIKEY, APPWRITE_ENDPOINT, APPWRITE_PROJECT_ID, and APPWRITE_API_KEY. Optional environment variables are GUARDIANKEY, SENTRY_DSN, MIN_BODY_CHARS, REQUIRE_IMAGE, CONTENT_RETENTION_DAYS, CONTENT_RETENTION_MAX_DELETES, and CONTENT_BODY_MAX_CHARS.

Output:

• On success, JSON with inserted, skipped_duplicate, skipped_empty, skipped_invalid_url, skipped_short_body, skipped_missing_image, insert_errors, insert_error_samples, duplicate_report, quality_report, and retention_report.
• On missing required environment variables, JSON with an error string.
• On uncaught error, JSON with error, inserted: 0, skipped_duplicate: 0, skipped_empty: 0, and skipped_invalid_url: 0.

Implementation notes:

• Fetches NewsAPI articles for en and de across category keywords.
• Fetches Guardian articles when GUARDIANKEY is set.
• Normalizes URL, title, body, summary, source, author, image URL, publication date, category, language, read time, and tags.
• Deduplicates incoming rows by exact sourceUrl and checks existing content rows by sourceUrl.
• Uses direct TablesDB-compatible REST fallback when SDK collection calls return 404.

fact_check

Purpose: validates a claim, reuses a recent exact cache hit when available, otherwise checks the claim with Perplexity and stores a fact-check row.

Field	Value
Runtime	node-22
Entrypoint	functions/fact_check/src/index.js
Timeout	30 seconds
Execute	any
Schedule	none
Install command	npm install
Build/runtime specification	s-0.5vcpu-512mb

Scopes:

• tables.read
• tables.write
• rows.read
• rows.write

Expected input:

{
  "claim": "Claim text, 500 characters or fewer",
  "userId": "optional user id for private history only"
}

The function supports OPTIONS for CORS and only accepts POST for execution. PERPLEXITY_API_KEY is required when no cache hit is returned.

Output:

• Success: JSON with claim, verdict, confidence, explanation, citations, and createdAt. Shared cache responses are anonymous and do not include userId.
• 400: { "error": "Claim is required and must be 500 characters or fewer" } or invalid JSON error.
• 402: { "error": "Insufficient API credits" } when Perplexity returns 402.
• 504: { "error": "Perplexity request timed out" }.
• 405: { "error": "Only POST requests are supported" }.
• 500: provider or internal error.

Implementation notes:

• Cache TTL is seven days.
• Cache lookup first tries exact equal("claim"), then falls back to search("claim").
• The Perplexity request uses model sonar and asks for JSON fields verdict, confidence, explanation, and citations.
• Shared fact_checks cache rows do not store caller userId; per-user history belongs only in user_fact_check_history.
• Primary storage writes newer fields (confidence, explanation, citations, createdAt) and legacy fields consumed by Flutter (confidenceLevel, confidenceScore, source, sourceUrl, contextBlock, checkedAt, language).
• On storage failure, it can return an unstored result.

chat_write

Purpose: performs trusted server-side writes for 1:1 threads, text messages, and article shares after verifying authenticated user and accepted friendship.

Field	Value
Runtime	node-22
Entrypoint	functions/chat_write/src/index.js
Timeout	30 seconds
Execute	users
Schedule	none
Install command	npm install
Build/runtime specification	s-0.5vcpu-512mb

Scopes:

• tables.read
• tables.write
• rows.read
• rows.write

Expected input by action:

{ "action": "getOrCreateThread", "otherUserId": "user id" }

{ "action": "postMessage", "threadId": "thread id", "body": "encrypted message body" }

{
  "action": "shareArticle",
  "recipientId": "user id",
  "contentId": "content row id",
  "message": "optional encrypted message"
}

The function reads the current user from Appwrite execution headers such as x-appwrite-user-id. It supports OPTIONS for CORS and only accepts POST.

Output:

• getOrCreateThread: { "thread": { ...thread row... } }.
• postMessage: { "messageId": "...", "threadId": "...", "sentAt": "ISO datetime" }.
• shareArticle: { "thread": { ...thread row... }, "messageId": "...", "sentAt": "ISO datetime" }.
• 401: { "error": "Sign in to continue." }.
• 403: { "error": "Only accepted friends can chat." } or thread access error.
• 400: missing field, invalid self-chat, too-long body/message, unsupported action, or invalid JSON.
• 405: unsupported method.
• 500: internal error.

Implementation notes:

• Text messages are limited to 4096 characters.
• Share messages are limited to 1000 characters.
• Threads are keyed by sorted participant IDs joined into pairKey; contentId is also written as dm:{pairKey} for compatibility.
• Thread, message, and share rows receive participant row permissions where row security applies.
• Typed article messages use type: "article" and contentId; a legacy body prefix is used if the typed schema fails.

like_sync

Purpose: synchronizes the denormalized content.likeCount after a client like row is created or deleted.

Field	Value
Runtime	node-22
Entrypoint	functions/like_sync/src/main.js
Timeout	15 seconds
Execute	users
Schedule	none
Install command	npm install
Build/runtime specification	s-0.5vcpu-512mb

Scopes:

• databases.read
• databases.write
• documents.read
• documents.write
• tables.read
• tables.write
• rows.read
• rows.write

Expected input:

{
  "contentId": "content row id, at most 64 characters",
  "liked": true
}

Output:

• Success: { "success": true, "likeCount": number }.
• 400: invalid liked state, invalid contentId, or invalid JSON.
• 404: { "error": "Content not found" }.
• 405: { "error": "Only POST requests are supported" }.
• 500: internal error.

Implementation notes:

• The function treats client input as the target liked state, reconciles the caller’s row in content_likes, recomputes the canonical count from content_likes, and writes that count back to the content row.
• Legacy increment / decrement requests are accepted for compatibility but are reconciled idempotently instead of blindly applying a numeric delta.
• It requires Appwrite endpoint, project ID, and API key from function context headers or environment variables.