Topic
Architecture and migration
Where a gateway sits, what it should and should not abstract, build-versus-buy, and moving an existing application without a rewrite.
18 articles · Last updated
Start with
More in this topic
Architecture & migration
AI model deprecation calendar: OpenAI, Anthropic, Google
One dated table of the model retirements OpenAI, Anthropic and Google have announced between now and mid-2027, with each vendor's listed replacement and a link to the page it came from, plus the provider-hosted copies that retire on a different day. Reviewed monthly.
Guide9 min read
Architecture & migration
Classifiers in the LLM request path: route, guard, monitor
A classifier in front of a model call can pick the model, refuse an injection or order the context; one behind it can score the answer. Each one adds a call that can be slow, wrong or down. Here is how to place them, budget them and decide what happens when they fail.
Guide11 min read
Architecture & migration
Manage your AI gateway from Claude Code or Cursor via MCP
An assistant that can read your spend, failures and routing is useful; one that can change them needs the same discipline as any other operator. This guide covers read versus write tools, token scopes, confirmation and audit, then how the INFRO MCP server applies them.
Guide10 min read
Architecture & migration
OpenAI's October 23 shutdowns: find every call before it fails
OpenAI's largest single-day retirement this year lands on 23 October: the o1, o3-mini and o4-mini reasoning models, the GPT-4 era snapshots, gpt-image-1 and every legacy fine-tune built on them. Here is the exact list, what changes for requests, and a ten-day plan to find every caller and move it.
Guide11 min read
Architecture & migration
Who owns your AI gateway? The 2026 acquisitions
Five well-known gateway and observability projects changed owner, status or name in 2026. None of that is a verdict on the products. It is a reason to know, for whatever sits in your request path, what you would have to move.
Analysis12 min read
Architecture & migration
Jev use cases: where typed AI decisions fit in a company
Most AI work inside a company is not writing. It is deciding: which queue, which policy, which model, which passage. These are the places TypeSafe's Jev fits, how each request is shaped, and where a person should stay in the loop.
Guide8 min read
Architecture & migration
From direct OpenAI calls to a gateway: a canary runbook
Changing the path every model call takes is a production change like any other. This runbook moves OpenAI traffic onto a gateway cohort by cohort, with the signals to compare at each stage and a rollback that is one configuration value.
Guide9 min read
Architecture & migration
Azure OpenAI and Bedrock behind one gateway: ids and quotas
Azure OpenAI and Amazon Bedrock both speak an OpenAI-style API, and almost nothing else about them lines up. This guide maps names, credentials, quotas and processing locations across the two, then puts them behind one route.
Guide8 min read
Architecture & migration
Multi-region AI traffic: latency, quotas and data location
Multi-region means four different things for AI traffic: surviving a regional outage, finding more quota, answering faster, and keeping data in a jurisdiction. They pull in different directions. This guide separates them and shows which control each one needs.
Guide8 min read
Architecture & migration
Self-hosted models behind an OpenAI-compatible endpoint
vLLM gives an open-weight model the same API shape as a hosted provider. Getting it into production means more than vllm serve: a stable model name, authentication that covers the right paths, a TLS edge, tool-call parsing, and a route that falls back when the GPUs are busy.
Guide8 min read
Architecture & migration
How to migrate to an OpenAI-compatible multi-model API
The base URL swap is the easy part. A safe migration also inventories provider assumptions, verifies response behavior, canaries traffic, and keeps rollback instant.
BriefUpdated 2 min read
Architecture & migration
AI infrastructure for startups: what to build and what to buy
Early teams should own product logic and evaluations, not a growing pile of provider adapters, retry loops, price tables, and invoice spreadsheets.
BriefUpdated 2 min read
Architecture & migration
Build vs buy an LLM gateway: the real engineering tradeoff
A proxy is a weekend. A gateway is an operating responsibility. Compare the long tail—streaming, tools, health, metering, policy, and maintenance—before deciding.
BriefUpdated 1 min read
Architecture & migration
How to avoid LLM vendor lock-in without limiting your product
Vendor independence does not mean avoiding every useful feature. It means isolating dependencies, measuring substitutes, and keeping model choice reversible.
BriefUpdated 1 min read
Architecture & migration
Multi-model API architecture: design one stable AI boundary
The goal is not to make every model identical. It is to give applications one stable boundary while capabilities and provider differences remain explicit.
BriefUpdated 1 min read
Architecture & migration
From one model to a multi-model AI platform: a staged plan
Do not begin with an elaborate router. Add one stable boundary, a measured substitute, clear traces, and policy—then let real failure modes justify each next step.
BriefUpdated 1 min read
Architecture & migration
OpenRouter alternatives in 2026: how AI gateway models differ
The useful comparison between AI gateways is not model count. It is who holds the provider accounts, what happens when your own key fails, whether budgets see that spend, what any fee is charged on, and who operates the thing in your request path.
ComparisonUpdated 8 min read