oh-my-common

Shared runtime for the oh-my-* Flask services. Extracted so each split-out service (oh-my-rss / oh-my-msg / oh-my-write) shares one implementation instead of drifting copies. Per the split design, each service keeps its own settings DB — this package is the code layer, not a shared store.

Modules

• settings — SQLite settings client + secret resolution. get_setting/put_setting/get_all, the provider_tokens cache (get/set/clear_cached_token), and resolve_secret(key, env_key) with precedence service-namespace > global > env. Convenience: resolve_claude_token(), resolve_openai_key(). Config via env: SETTINGS_DB (path), OMI_SERVICE_NS (this service's namespace, e.g. clipboard).
• claude — Claude CLI gateway: resolve_claude_bin, claude_text, run_claude_capture, gate_ai, claude_status, extract_json, estimate_tokens, estimate_cost, MODEL_PRICING. Flask-free.
• tz — local_tz()/now_local()/today_local() from OMI_TZ_OFFSET (hours, default +8).
• urls — request_base_url() (honors X-Forwarded-Proto/Host). Imports flask lazily.
• scheduler — make_scheduler(ENABLED_ENV) → a guarded BackgroundScheduler (returns None when ENABLED_ENV=0, so extra gunicorn workers don't double-run). Imports apscheduler lazily.
• llm — complete(prompt, model='sonnet', backend='claude_cli'): backend-routed one-shot completion. Only claude_cli (the claude -p gateway) is wired today; Anthropic/OpenAI HTTP backends are a future add (the seam is in place). claude_text stays as the claude-cli shortcut.
• agent — dify_run(inputs, query, ...): shared client for a self-hosted Dify app (blocking). Creds (dify_url/dify_key) from per-service settings or args. Returns {connected, result|error}. Consolidates the hand-rolled Dify calls that lived in service route code.

Usage

import os
os.environ.setdefault('OMI_SERVICE_NS', 'clipboard')
from oh_my_common import get_setting, claude_text, resolve_openai_key, now_local

key, src = resolve_openai_key()
text = claude_text('Summarize: ...', model='sonnet')

Install

pip install oh-my-common            # core (stdlib only)
pip install 'oh-my-common[flask,scheduler]'   # + request_base_url + make_scheduler

Distribution into Docker services is TBD — see the migration plan. Options: publish to a registry, vendor into each monorepo, or a git dependency.