Hidden Costs of AI APIs: Context History and Rate Limits Add Up Fast

·2 views

A developer who spent six months building an AI analytics tool found that the per-token API price was only part of the expense. Chatbots resend the entire conversation history with every request, so a short question can balloon into thousands of billed tokens. Implementing a context summarizer cut costs by about 60%, though it added engineering complexity. Rate limits designed for prototyping also caused 429 errors during a client demo, requiring custom retry-with-backoff logic. The author argues that developers should budget for these overlooked costs, including the engineering time needed to handle them.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Connecting your own Supabase project to Lovable apps: setup and key caveats

Lovable apps get Lovable Cloud by default, which runs on Supabase, but this guide focuses on connecting a Supabase project you own, which is typically preferred for apps with real users. Setup happens in two steps: a workspace owner or admin first links a Supabase organization, then an editor connects a Lovable project to one of its Supabase projects. A common snag is that editors cannot add an organization themselves, so the owner or an admin must complete that step first. Switching between Cloud and your own Supabase later is not one-click and requires manual migration work, so it is best decided at the start. Once connected, schema changes are saved as migration files in the repository, while data changes and secrets stay under your control in the Supabase dashboard, and the author stresses manually verifying row-level security policies before launch.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Browser-Based Image-to-Bead Pattern Generator Uses CIE Lab Color Matching

A developer team built BeadPattern, a tool that converts photos into fuse bead patterns entirely within the user's browser, so images never reach a server. Matching colors in standard RGB produced muddy results for pastels and skin tones, so the team converted pixels and bead palettes into CIE L*a*b* space and used CIE76 distance instead. To keep slider adjustments responsive, they precomputed palette vectors and indexed them in a K-d tree for faster nearest-neighbor lookups. They also applied Floyd-Steinberg error diffusion to reduce color banding in gradients when mapping to a limited palette.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Netflix Q3 2026 earnings prep: guidance history, events and deck via MCP server

A developer who builds the Equibles stock market data MCP server shows how an AI agent can assemble investor relations material for Netflix's Q3 2026 earnings interview, scheduled for 20 October at 20:45 UTC. The tool lists the company's investor events, including the earnings call, the June annual meeting and a March conference appearance, with transcripts and slides where available. Tracking Netflix's full-year 2026 guidance across three releases shows the revenue range narrowing in July, the operating margin target unchanged at 31.5%, and the free cash flow target raised by $1.5B in April. The author also describes how a CFO's March conference remarks can be checked against the January plan, and notes the tool flags releases that have not yet been fully processed.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Gemini Gas Optimization Audit Finds Savings Potential and Gas-Related DoS Risks

An audit of Gemini's on-chain contracts, which hold roughly $5.34 billion in total value locked across Ethereum and Layer 2 networks, examined gas efficiency and gas-related attack vectors. Reviewers used static analysis, mainnet-fork profiling, and L2 testing on Arbitrum and Optimism to identify inefficiencies. Key issues included storage-slot fragmentation, quadratic gas growth in batch settlement, and external calls without gas caps that could enable griefing on congested L2 networks. The auditors estimated optimizations could cut gas costs by about 12 to 22 percent per operation, potentially saving around $3.2 million annually. The overall risk score was rated 4 out of 10, with recommended upgrades targeted within four to six weeks.

0 comments Read more at DEV Community