Halv is open source
The desktop app and its local savings engine now use the Apache 2.0 license. Halv Cloud, the part that runs on our servers, is the only paid part. Here is why.
Read postHALV BLOG
Product updates.
Engineering notes.
News about Halv and notes from the team building it.
Claude Code vs Codex: Which Fits Your Workflow?
Compare Claude Code and Codex for terminal work, cloud tasks, permissions, and usage limits. Choose with a practical test on your own repository.
Read postHow to Reduce ChatGPT Usage
Reduce avoidable ChatGPT usage with focused chats, relevant files, controlled tools, shorter outputs, and the right workspace for coding.
Read postHow to Reduce Claude Usage
Reduce Claude and Claude Code usage with smaller context, focused Projects, clear model choices, controlled tools, and complete measurement.
Read postHow to Reduce Codex Usage
Reduce Codex plan usage with tighter context, scoped tools, suitable models, measured workflows, and task-aware routing.
Read postSame pass count. 57.1% lower model cost.
Halv and vanilla Codex each passed 25 of 42 benchmark runs. Halv recorded $144.67 in model cost versus $337.50. Here is the result in plain language.
Read postInside Halv’s 57.1% cost reduction: the SWE-rebench data
Across 14 tasks and three repetitions each, Halv and vanilla Codex both passed 25 of 42 runs. Halv recorded $144.67 in model cost versus $337.50.
Read postAGENTS.md Token Usage: How Much Do Instructions Cost?
Audit AGENTS.md without losing project rules. Learn how Codex loads instructions, why bytes differ from tokens, and how to verify a shorter file.
Read postCodex /compact vs /new: Which Saves More Tokens?
Choose between Codex /compact and /new. Learn what each keeps, when a reset creates repeated work, and how to compare total task usage.
Read postHow Much Context Do MCP Servers Add? A Token Usage Guide
Understand MCP token usage in Codex and Claude Code. Separate tool definitions, discovery, and results, then measure the cost of useful work.
Read postWhy Is Codex Using So Many Tokens? 7 Things to Check
Find what drives Codex token usage: context, repeated reads, reasoning, tools, and subagents. Use a practical checklist before changing your plan.
Read postHalv used 51% fewer tokens per correct answer on SWE-rebench
On 20 paired SWE-rebench tasks, Halv used fewer tokens, cost less, and solved more tasks than vanilla Codex under the same model and verifier.
Read postHow to save tokens in ChatGPT and Codex
Reduce avoidable ChatGPT and Codex usage with smaller context, scoped instructions, suitable models, lower effort, and measured results.
Read postHow to save tokens in Claude Code
Reduce avoidable Claude Code usage with smaller context, focused instructions, suitable models, lower effort, and complete workflow measurement.
Read postCodex usage limits: resets, plans, and credits
Learn how Codex tokens, credits, usage limits, ChatGPT subscriptions, and API billing differ.
Read postClaude usage limits: resets, plans, and credits
Learn how Claude session limits, weekly limits, context length, usage credits, API costs, and subscriptions differ.
Read postCrux in real sessions: 90% correct, 47% cheaper per answer
The follow-up benchmark: Crux in real multi-question coding sessions on django and sympy. 90% correct answers versus grep's 58% — at 47% less cost per correct answer.
Read postCrux in sessions: the benchmark that forced a result-format redesign
Methodology and full data for the session-shaped Crux benchmark, the 2.3x token blowup it exposed on sympy, and the Crux 0.6.2 output redesign that turned it into a 59% saving.
Read postOne of Halv's tools made AI coding agents 45% more accurate
We benchmarked an AI coding agent with and without Crux, our code-index MCP server, in Codex on django and sympy: 45% more correct answers, 24% less cost per answer.
Read postCrux vs grep: benchmarking an SCIP code index for AI coding agents
Methodology and full data for the Crux benchmark: an SCIP code-index MCP server vs grep in Codex CLI, with token-metered paired runs, ablations, and the tuning trail.
Read post