RouterLane
Changelog

What changed

New first.

  1. RouterLane: a new name and a new home

    • The service is now RouterLane, at routerlane.com.
    • The API lives at https://api.routerlane.com, with /v1 for OpenAI clients.
    • New keys start with rl_. Existing keys keep working.
    • Model IDs accept the routerlane/ prefix, for example routerlane/smart.

  2. Smart+

    • A fourth tier, smart-plus: the client uploads its repository once and the job runs on RouterLane's machine.
    • Setup, explore, build, test and an independent QA review, with repair rounds when tests fail or QA finds blocking issues.
    • One patch back, applied with git apply. Nothing is committed.
    • A per-job spend ceiling, $1.50 by default. Workspaces are deleted after 48 hours.
    • Measured: a 5-file package in 24 s for $0.035, and RouterLane's own 45-file repository in 2.3 min for $0.75.

    How Smart+ works

  3. Insights: ask the data, session summaries, content filters

    • Ask the data: pick users, sessions and a time window, ask a question, and get an answer that cites turns and sessions.
    • Session summaries: title, goal, status, progress, problems, next step and user mood, kept up to date as a session grows.
    • Content filters: yes or no rules over prompts, replies, reasoning, tool calls and tool results, with log or block actions.
    • Block mode: a blocked input is never forwarded and returns 403; a blocked reply is cut and the client gets an error event.

    The console

  4. Transcript explorer

    • Every session organized as user, session, thread, turn and step.
    • Views for Conversation, Prompts and answers, Reasoning, Tool calls, Routing and Raw.
    • Full-text search, and filters by user, app, tier, model, complexity, kind, decision, tool and errors.
    • Logging settings: transcripts on or off, raw parameters on or off, retention 30 days by default.

  5. Tiers, routing and the benchmark

    • Three tiers: fast, normal and smart.
    • A small decision model reads each new task in 200 to 350 ms and classifies its complexity and kind; per-tier rules pick the model and effort.
    • Three API formats: Anthropic Messages, OpenAI Chat Completions and OpenAI Responses.
    • Failover before the first byte, circuits that open after three failures in a row, and a queue that serves Fast first.
    • A benchmark of 13 graded tasks per model and effort, which set the default rules.

    How routing works