blindxfish/devplacepy
1
0
Fork 0
forked from retoor/devplacepy
Code Pull Requests Activity

feat: add AI markdown reformatting and usage metering to NewsService

Browse Source 
Add AI-powered body reformatting for valid articles in the news pipeline, converting raw text walls into clean Markdown with paragraphs, headings, and lists. Introduce `news_usage` database table and `add_news_usage`/`get_news_usage` helpers to track per-cycle gateway costs (calls, tokens, latency, USD) from response headers, reported on the admin Services page. Remove unused `.sidebar-more` CSS and apply `render_title()` to poll question/label fields.
This commit is contained in:
retoor
2026-06-18 23:46:53 +00:00
parent 6ceca3d0d4
commit f3a4667fce
13 changed files with 427 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CLAUDE.md
+1 -1
View File
@@ -165,7 +165,7 @@ The `background` singleton is a generic, lightweight, fire-and-forget offload fo
A sibling of AI content correction that reuses its plumbing (`CORRECTABLE_FIELDS`, `PENDING_SCOPE_KEY`, `gateway_complete`, the sync/background apply-mode mechanics, the per-user usage upsert) and differs only in two ways: it is **enabled by default** and **synchronous by default**, and it runs ONLY where an authored prose field contains an inline `@ai <instruction>` directive (`has_ai_directive`, regex `@ai\s+\S`). When present, the configured prompt (default `config.DEFAULT_MODIFIER_PROMPT` = "Execute what is behind `@ai` (the prompt) and replace that part including `@ai`") tells the model to execute the instruction and replace the marked part including the `@ai` marker; triggerless writes cost nothing. **It is context-aware:** `_run_modification` builds a grounding **context block** once per row via `services/ai_context.py` `build_context(table, uid, row, user_uid)` and passes it to `modify_text`, which appends it to the system message. The block (fail-soft, length-capped) carries the date, the author's stats (username/role/level/stars/posts/rank/followers/member-since/bio), and table-specific location/thread context (a comment's target post + parent reply; a post's topic/project; a project/gist's title/description, and a gist's language + read-only `source_code`; a DM's recipient + recent conversation), so `@ai answer the question above` / `@ai write my bio from my stats` / `@ai reply to this` work. `schedule_modification(user, table, uid, request)` is hooked alongside `schedule_correction` at the same content/comment/messaging/profile entrypoints. Settings are three `users` columns (`ai_modifier_enabled`/`ai_modifier_sync`/`ai_modifier_prompt`, seeded born-live in `utils._create_account`), edited at the owner-or-admin leaf `POST /profile/{username}/ai-modifier` (`routers/profile/ai_modifier.py`, `AiModifierForm`, audit `profile.ai_modifier`), surfaced owner-only on the profile (`ai_modifier_*` + `modifier_usage` on context + `ProfileOut`, dollar cost admin-only), wired by `static/js/AiModifier.js` (`app.aiModifier`), and driven via the Devii `ai_modifier_get`/`ai_modifier_set` tools (`services/devii/ai_modifier/`, `handler="ai_modifier"`). Per-user totals accumulate via `database.add_modifier_usage(user_uid, totals)` into the `modifier_usage` table (same SUMS-plus-latency columns and computed performance averages as correction, shown on the profile with dollar averages admin-only). See `AGENTS.md` -> "AI modifier".
### Background services (`devplacepy/services/`)
`BaseService` provides the async run loop, a `deque(maxlen=20)` log buffer, and graceful cancellation. `ServiceManager` is a singleton that registers, starts, and stops services. `NewsService` (`services/news.py`, `default_enabled=True`) is a fully automatic, zero-maintenance import pipeline: it fetches articles from `news_api_url`, **cleans** each (HTML strip + `clean_news_text`/`JUNK_PATTERNS` Reddit-boilerplate removal), **fetches and perceptually compares the images** (SSRF-guarded fetch + Pillow decode + `imagehash` phash off-thread; placeholder = too small / undecodable / a phash shared across 2+ different articles), **grades deterministically** (AI grade on cleaned text via `news_ai_url` + a reliability gate + a unique-image bonus and thin-content penalty -> effective `grade`, raw in `ai_grade`), and inserts ALL of them into `news` with `status="published"`/`"draft"` on `news_grade_threshold` - nothing is silently skipped. After the loop it **auto-rotates Featured + Landing** (`_apply_landing_selection`, top scored unique-image articles in the recent window) while honouring per-row `featured_locked`/`landing_locked` set when an admin manually toggles. The full ruleset is on `NewsService.description` and surfaces on `/admin/services`, which also polls live status and the log tail.
`BaseService` provides the async run loop, a `deque(maxlen=20)` log buffer, and graceful cancellation. `ServiceManager` is a singleton that registers, starts, and stops services. `NewsService` (`services/news.py`, `default_enabled=True`) is a fully automatic, zero-maintenance import pipeline: it fetches articles from `news_api_url`, **cleans** each (HTML strip + `clean_news_text`/`JUNK_PATTERNS` Reddit-boilerplate removal), **fetches and perceptually compares the images** (SSRF-guarded fetch + Pillow decode + `imagehash` phash off-thread; placeholder = too small / undecodable / a phash shared across 2+ different articles), **grades deterministically** (AI grade on cleaned text via `news_ai_url` + a reliability gate + a unique-image bonus and thin-content penalty -> effective `grade`, raw in `ai_grade`), **reformats every valid article into clean Markdown** (`_format_article` + the editable `news_format_prompt`; paragraphs, `## ` headings, lists and code, preserving every fact; fail-soft to the cleaned original, toggle `news_format_enabled`), and inserts ALL of them into `news` with `status="published"`/`"draft"` on `news_grade_threshold` - nothing is silently skipped. After the loop it **auto-rotates Featured + Landing** (`_apply_landing_selection`, top scored unique-image articles in the recent window) while honouring per-row `featured_locked`/`landing_locked` set when an admin manually toggles. **AI usage is metered like the correction/modifier consumers**: each gateway call's `X-Gateway-*` response headers are parsed (`parse_usage_headers`) and the run's totals accumulated into the durable single-row `news_usage` table (`database.add_news_usage`/`get_news_usage`, same SUMs-plus-computed-averages shape as `correction_usage`); `NewsService.collect_metrics()` surfaces calls/tokens/cost and the per-call averages as `stats` on the admin-only `/admin/services` page. The full ruleset is on `NewsService.description` and surfaces on `/admin/services`, which also polls live status and the log tail.
`GatewayService` (`/openai/v1/*`) is the **single point of truth for AI**. Every other AI consumer (news, bots, Devii guests) defaults its endpoint to `config.INTERNAL_GATEWAY_URL` (`http://localhost:10500/openai/v1/chat/completions`, override via `DEVPLACE_INTERNAL_BASE_URL`), sends the generic model `molodetz`, and authenticates with the auto-generated `gateway_internal_key` (via `database.internal_gateway_key()`) when its own key is unset. **Per-user attribution:** the gateway also accepts a real user's `api_key` (Bearer / X-API-KEY / session), gated by `gateway_allow_users` (default on); `resolve_owner()` records `(owner_kind, owner_id)` per call, so usage is attributed and limitable per user. Devii operating a signed-in user authenticates its LLM calls with that user's own `api_key` (set as the session's `ai_key` in `build_settings(..., owner_kind="user")`), so a user's full gateway spend (Devii and direct API calls) rolls up under their uid - surfaced admin-only on the profile page via `build_user_usage()` / `GET /admin/users/{uid}/ai-usage`. Guests keep the internal key. **System-message composition (`services/openai_gateway/system_message.py`, `handle_chat` only):** before assembling the upstream chat payload (after vision augmentation, so it reaches every chat consumer) the gateway composes ONE system message in the fixed order **operator preamble -> date line -> client system content**. (1) The admin-editable `gateway_system_preamble` config field (Prompt group, multiline text, default empty) is prepended ahead of the client's system content on every chat call; if the client sent no system message it becomes a new leading one (a blank preamble is a no-op). (2) The **current date in EU `DD/MM/YYYY` (date only, never time** - omitting time keeps the message byte-stable per day so upstream prompt caching stays effective) is injected only when the composed text has NO date in any common format, detected by the `contains_date` regex set (ISO `YYYY-MM-DD`, `D/M/YYYY`, `DD-MM-YYYY`, `DD.MM.YYYY`, `MM/DD/YYYY`, and textual months like `14 June 2026`/`June 14, 2026`/`14 Jun 2026`). Embeddings/passthrough are untouched. Real provider keys/URLs/models live only in the gateway; `database.migrate_ai_gateway_settings()` (run in `init_db()`) generates the internal key, migrates `DEEPSEEK_API_KEY`/`OPENROUTER_API_KEY` env into the editable, non-secret `gateway_api_key`/`gateway_vision_key` settings, and rewrites old external AI URLs to the gateway. The gateway is `default_enabled=True`. The gateway also serves OpenAI-compatible text **embeddings** at `POST /openai/v1/embeddings`: clients send the client model `molodetz~embed`, which `handle_embeddings` remaps to the configured embedding model (default `qwen/qwen3-embedding-8b` on OpenRouter, $0.01/1M input tokens), tracked per call as `backend="embed"` in `gateway_usage_ledger`. The gateway records every upstream call to `gateway_usage_ledger` (cost from the upstream native `cost` field when present, else computed from per-1M pricing) and samples `gateway_concurrency_samples`; `services/openai_gateway/analytics.py` rolls these into per-hour/24h token, cost, latency, error, and behavior metrics shown on `/admin/ai-usage` (`AiUsageMonitor.js`, schema `GatewayUsageOut`, Devii action `ai_usage`). It also retries transient upstream failures and trips a circuit breaker (`services/openai_gateway/reliability.py`). TTFT/inter-token latency are not tracked because the upstream is called non-streaming. **Every gateway response also carries per-call `X-Gateway-*` headers** (model, full token breakdown, `Cost-USD`/input/output dollars, native-cost flag, tokens/sec, latency, context window/utilization): `GatewayUsageLedger.record` returns the computed ledger row and each handler's `finalize` maps it through `usage.usage_response_headers` onto the `Response`, so any caller can read its own spend (the AI-correction worker uses this).