News & guides
Context Compaction: Keeping Agents Alive Past Turn Twenty
Agent transcripts grow until quality degrades and cost climbs. What to summarise, what to drop, and what must never be compacted away.
· 9 min read
Context Length vs Effective Context: The Number You Can Use
A model advertising a one-million-token window does not reliably use one million tokens. The gap between the spec sheet and what actually works.
· 9 min read
Context Windows Compared: Input, Output and What It Costs
Filling a million-token window costs between fourteen cents and three dollars depending on the model. The full comparison, including output ceilings.
· 9 min read
Continuous Batching: How Servers Keep GPUs Busy
Continuous batching lets finished requests leave a batch and new ones join mid-flight. It is why modern inference servers hold high load without stalling.
· 9 min read
Cost Per 1K Lines of Code: A Tempting, Mostly Bad Metric
Lines of code is the easiest denominator for AI spend and one of the worst. Where it misleads, where it genuinely works, and what to measure instead.
· 9 min read
Cost Per Agent Run: Why Input Dominates the Bill
Agent costs are driven by resent transcript, not generated output. Working out what one run actually costs and which lever moves it.
· 9 min read
Cost Per Code Review: What Automated Review Actually Costs
Automated review looks cheap per pull request until you count re-reviews, large diffs and false positives. Working out the real per-review figure.
· 8 min readCost Per Refactor: Estimating a Multi-File Change Up Front
Refactors are the worst case for context bloat because every touched file must stay in view. A method for estimating the bill before you start the run.
· 10 min read
Cost Per Test Suite: Writing Tests Versus Fixing Them
Generating tests with a model is cheap and bounded. Making a red suite green is neither. How the two costs differ and how to put a ceiling on the expensive one.
· 10 min read
Cost-Adjusted Model Scoring: Ranking by Value, Not Capability
Capability leaderboards ignore price entirely. How to build a score that reflects what a model costs to run on your workload, and where it misleads.
· 9 min readCritic Agent Patterns: Useful Review Without Endless Rewrites
A critic agent reviews work another agent produced. The patterns that improve output, the ones that just add cost, and how to tell them apart.
· 9 min read
curl Recipes for LLM APIs: Debug Before You Write Code
A working set of curl commands for OpenAI-compatible endpoints: streaming, timing, tool calls, error bodies, and building JSON safely with jq.
· 9 min readReady to try it?
Every model behind one key. Start free and upgrade when it sticks.