Heuristic API-Proxy for Off-Peak Routing (HAP-OR)

Who it's for

Small SaaS startups and indie developers with high background LLM workloads (e.g., daily vector ingestion, email summarization, bulk categorization) who currently pay standard on-demand API rates.

The problem

Payment. LLM API bills are unsustainable for high-volume background tasks, and manual cron/queuing implementations to exploit off-peak pricing require complex infrastructure maintenance.

How to build it

A lightweight, self-hosted Docker proxy running on the user's infrastructure, configured via environment variables to swap standard API endpoints.

How it makes money

SaaS builders pay $15/month for a premium, pre-built binary/dashboard with advanced rate-limit modeling, or $49/month for a hosted relay version to avoid managing docker containers.

Why it doesn't exist yet

Incumbents like OpenAI or DeepSeek focus on Selling raw compute and dynamic demand-based pricing models; they have no financial incentive to build client-side, cross-provider scheduling optimization that reduces their own peak-load margin. Indea builders can easily ship a drop-in proxy because it requires zero modification of the existing codebase beyond the base URL.

First users

Indie developers who run background processing tasks (such as batch newsletter analysis or vector syncs) and are hit with sudden $200+ monthly API bills.

Build size

1 person x 4 weeks. Includes an outbound API gateway proxy, an SQLite-backed local queue, a config dashboard to define 'urgency levels' per payload, and scheduled cron integrations for DeepSeek/OpenAI off-peak discount hours.

Biggest risk

The risk of major providers completely eliminating peak/off-peak pricing differentials is very low, as cloud grids and AI data centers inherently rely on load-shifting to manage power and compute capacity.

Conditions for a hit (all 3 required)

• Interception and parsing of outgoing LLM requests to inspect a custom header ('X-Urgency: low') indicating the request can be delayed for off-peak pricing.
• An automated local SQLite queue that holds low-priority payloads and dispatches them instantly when provider-specific discount windows begin.
• A local diagnostic UI showing real-time API cost savings accrued over 30 days based on active provider pricing tables.

How it's judged (in 6 months)

GitHub repository with over 400 stars or self-hosted deployment launch reaching top 10 on Product Hunt within 6 months.(judgment date 2027-02-15)

AI self-confidence 78/100 — self-reported likelihood of meeting the criterion, not a business success rate