Building RouteAPI. Practical notes on AI APIs, model routing, and production workflows. Personal views.
Joined June 2026
- Tweets93
- Following27
- Followers3
- Likes2
Per-user keys fix audit but multiply the model-parameter config you mentioned. Would your security team accept one org-managed key with per-user attribution, or is a separate key per developer a hard requirement?
Deleting a stored secret is not key revocation.
For affected credentials, revoke or rotate the key with its issuing provider, then update every application that used it. Removing the saved copy doesn't end its access.
Agreed, warm-up time is the line item most migration estimates skip. A few hours of cache misses at full input price can eat a big share of the per-token savings, so I'd measure it on a slice of traffic before switching.
Tracking spend per coding tool is useful. But it doesn't tell you which one finishes tasks efficiently.
If you're running Claude Code, Codex and OpenCode, I'd also want cost per completed fix.
How are you measuring that today?
Our Usage dashboard just got a major refresh!
New charts and visuals for:
- Spend & token tracking: Break down cost, tokens, and requests across models, virtual keys, teams, and users.
- Agent attribution: See usage by agent (Claude Code, Codex, OpenCode, and more)
- Targeted insights: Filter by date range or user, then drill down into model, key, and MCP server activity.
Some AI calls only need a yes/no, a choice, or a score.
OpenAI's Decisions API returns typed answers with probabilities. Input starts at $0.10/M tokens, with no output-token charge.
Which agent step would you move first?
Lowest price per prompt can still mean a higher session bill. Do routing presets account for cache misses when consecutive calls land on different providers?
That cache hit rate makes provider switching a different cost problem. Have you tested what one cold-cache day would do to the city's daily budget?
I'd pin a model per workflow until task-level evals justify a switch. Are you routing through one gateway, or managing provider clients separately?
Before tuning budgets, I'd check which models each subagent is allowed to call. Do they inherit every configured provider, or have their own allowlists?
Collapsing tool calls helps readability. I'd still want the summary to flag touched files and permission changes. Does it surface those?
For long coding loops, cost per completed task is usually more useful than token price alone.
Are you testing Spark and Flash on the same tasks, or switching based on the workload?
A common protocol gets you portability. State and tool behavior decide whether the switch is actually painless.
Are you using more than one coding harness today?
Coding-agent pricing has become a stack, not a single bill.
Some developers pay for Claude Code or Codex, then still keep API spend for automation.
What's your setup today: subscription, direct API, gateway—or a mix?
At $0.00035 per judgment, the interesting shift is coverage, not just savings.Online evals can move from sampling a few traces to scoring far more of them.Would you judge every production run at that price?
I'd isolate the gateway before touching both clients: one raw request, one key, one model ID.
Is the 401 coming from the gateway itself or the upstream provider?
Open-weight models handled 56% of Vercel AI Gateway tokens in August, but only 14% of spend.
That split matters more than another “best model” ranking.
Where are you already using open-weight models in production?
Free makes Union Alpha easy to test. The bigger production choice is whether you'd depend on a stealth provider.Would you ship it directly, keep it as fallback, or only use it for experiments?