fix: configurable AI request timeout with retry on timeout errors

[Copilot]

AI API calls used a hardcoded 120s timeout with no env var to override it, and timeout errors caused immediate failure with no retry — leaving users with large AI_MAX_OUTPUT_TOKENS or slow connections stuck in a dead end.

Changes

• New AI_REQUEST_TIMEOUT env var (default 300s, min 30s) — wired through config.py → AIConfig → AIRecommendationEngine.__init__
• Removed hardcoded timeout=120 in _generate_with_openai — now uses self.request_timeout
• Timeout retry support in _generate_with_retry — timeout errors now trigger exponential backoff retries (same as rate limits), detected via isinstance(e, (TimeoutError, ConnectionError)) + type-name/message matching for broad provider coverage (httpx, openai SDK, google-genai)
• Docs updated — ENV_VARS.md (new section + summary table), README.md quick-reference table, .env.example

# Increase if gpt-4o-mini or other slow models time out on large responses
AI_REQUEST_TIMEOUT=600

[Copilot]

Pull request overview

Adds a configurable AI API request timeout and extends the AI generation retry logic to also retry on timeout-like failures, improving resilience for slow models / large outputs.

Changes:

• Introduces AI_REQUEST_TIMEOUT (default 300s, min 30s) in config/models/docs and passes it into AIRecommendationEngine.
• Removes the hardcoded OpenAI request timeout and uses the configured timeout instead.
• Updates retry logic to treat timeout errors as retryable with exponential backoff (similar to rate limits).

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
README.md	Documents AI_REQUEST_TIMEOUT in the env var quick-reference table.
ENV_VARS.md	Adds full documentation for AI_REQUEST_TIMEOUT and updates the variable summary.
.env.example	Adds AI_REQUEST_TIMEOUT example entry.
octogen/models/config_models.py	Adds AIConfig.request_timeout with validation (min 30s).
octogen/config.py	Loads AI_REQUEST_TIMEOUT into the env-derived config dict.
octogen/main.py	Passes request_timeout into AIRecommendationEngine (but currently doesn’t load the env var in this entrypoint).
octogen/ai/engine.py	Uses self.request_timeout for OpenAI calls and retries on timeout-like errors.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

request_timeout is read from self.config['ai'], but _load_config_from_env() in this module never populates an ai.request_timeout key (and AI_REQUEST_TIMEOUT isn’t read anywhere in main.py). As a result, the new env var won’t actually be configurable via the main entrypoint and will always fall back to 300.

Wire AI_REQUEST_TIMEOUT into the ai config dict (e.g., via _get_env_int) and enforce the documented minimum (30s) in _validate_env_config() or when parsing the env var.

[Copilot]

is_timeout treats all ConnectionError exceptions as timeouts. ConnectionError also covers non-timeout failures (e.g., DNS resolution errors, connection refused, misconfigured AI_BASE_URL), so this can introduce long retry delays where an immediate failure would be more actionable.

Consider narrowing the type check to timeout-specific exceptions only (and/or gating connection errors on a timeout-related message/errno), or splitting “timeout” vs “connection” handling so only genuine timeouts get the exponential backoff retries.

[Blueion76]

@copilot apply changes based on the comments in this thread

[Copilot]

@copilot apply changes based on the comments in this thread

Both issues addressed in ea02085:

1. AI_REQUEST_TIMEOUT not wired — main.py has its own _load_config_from_env() that was missing the key. Added "request_timeout": max(30, self._get_env_int("AI_REQUEST_TIMEOUT", 300)) to the ai dict there, enforcing the documented 30s minimum.

2. ConnectionError too broad — Removed ConnectionError from the isinstance check. The timeout detection now relies only on isinstance(e, TimeoutError), 'Timeout' in error_type, and message phrases (timed out, timeout, etc.). This means genuine connection failures (DNS errors, connection refused, misconfigured AI_BASE_URL) will correctly fail immediately instead of being retried with backoff.