Built with Astro, hosted on Cloudflare. Domain registered in 2026.
halv.ai is built with Astro, hosted on Cloudflare. It measures traffic with Cloudflare Web Analytics. It sends email through Amazon SES.
| Technology | Version | First seen |
|---|---|---|
| JavaScript Frameworks & Libraries1 | ||
| 7.0.7 | 6 Nov 2026 | |
| Frontend Build Tools & CSS1 | ||
| – | 6 Nov 2026 | |
| CDN & Hosting1 | ||
| – | 6 Nov 2026 | |
| Analytics & Tag Management1 | ||
| – | 6 Nov 2026 | |
| Advertising & Affiliate Tracking1 | ||
| – | 6 Nov 2026 | |
| Transactional Email2 | ||
| – | 6 Nov 2026 | |
| Accessibility1 | ||
| – | 6 Nov 2026 | |
| Comments & Community1 | ||
| – | 6 Nov 2026 | |
| Fonts & Icon Fonts2 | ||
| – | 6 Nov 2026 | |
| Security, WAF & Bot Protection1 | ||
| – | 6 Nov 2026 | |
Registration from the registry's RDAP record, the certificate the site serves, and its rank in public traffic lists.
The crew hasn’t read this site’s colors yet. We read a palette when we capture a site’s screenshot, busiest sites first. Meanwhile, browse palettes from other sites.
Counted from every stack change we've recorded: of the sites we watched running one of this site's tools, how many went on to add another. A pattern across sites, not a forecast for this one.
route1.mx.cloudflare.netroute2.mx.cloudflare.netroute3.mx.cloudflare.netmina.ns.cloudflare.comgarret.ns.cloudflare.comgoogle-site-verification=I08AtUQ6kvSHoQ_vJFMNwAqXkGcZLLGeALcS9cZLLYIv=spf1 include:_spf.mx.cloudflare.net ~allUser-agent: * Allow: / Disallow: /internal-locale/ Disallow: /r/ Disallow: /i/ Sitemap: https://halv.ai/sitemap-index.xml
# Halv > The open-source desktop workspace for AI coding agents. Halv is an open-source desktop workspace for AI coding agents. Run Claude Code, Codex, Kimi, and GLM in one app with chat, native terminals, project history, and a live savings meter. Code navigation, command output filtering, and context compression run on your computer to reduce wasted tokens. The desktop app is free under Apache 2.0. Halv Cloud adds JEV routing for US$10 a month. Across 14 SWE-rebench tasks with 3 repetitions each, Halv recorded 57.1% lower model cost. Both workflows passed 25/42 runs. Halv cost $144.67; vanilla cost $337.50. Halv used 3.49 times as many total tokens. Halv published this interim workflow comparison. It covers the first 14 of 111 planned tasks. Equal aggregate pass counts do not imply identical solved tasks. The result does not establish Claude Code savings or independent endorsement by SWE-rebench. Infrastructure-invalid attempts and a rolled-back policy experiment are excluded. Some excluded costs are unknown. ## Product - [Product facts and complete FAQ](https://halv.ai/llms-full.txt): Plain-text product details, the open-source license, Halv Cloud pricing, data handling, and benchmark scope. - [Desktop workspace](https://halv.ai/): The open-source desktop app, Halv Cloud, JEV routing, benchmark proof, and a comparison with other tools. - [Source code](https://github.com/halv-ai/halv): The Halv desktop app source, licensed under Apache 2.0. - [Frequently asked questions](https://halv.ai/#faq): Answers about open source, Halv Cloud, routing, privacy, agents, and platforms. - [Task routing](https://halv.ai/#router): JEV routing, part of Halv Cloud. The halv-savings tool on your computer sends each brief to Halv servers, where JEV rates it. Halv then picks a lane, model, and reasoning effort for the task. - [Pricing](https://halv.ai/#cloud): The desktop app is free and open source. Halv Cloud costs US$10 per month. - [Download](https://halv.ai/download/): Official installers for macOS, Windows, and Linux. ## Claude Code and Codex usage - [How can I increase my Claude Code usage limit?](https://halv.ai/blog/claude-usage-limits-explained/): To make your current allowance last longer, reduce unnecessary context and tool output. Halv combines local context compression, code indexing, and command output filtering in one workspace. Anthropic still controls your account's limits. - [What is the best way to save tokens in Claude Code?](https://halv.ai/blog/how-to-save-tokens-in-claude/): Start by reducing unnecessary context: keep instructions focused, give Claude relevant files, and use /clear when starting unrelated work. Halv adds code indexing, command output filtering, and local context compression around Claude Code in a desktop workspace. Use its savings meter to measure the difference in your own work. Savings depend on the task and context. - [How can I increase my Codex usage limit?](https://halv.ai/blog/codex-usage-limits-explained/): To make your current Codex allowance last longer, keep requests focused and reduce unnecessary context. Halv helps reduce token waste with code indexing, command output filtering, and context compression. OpenAI controls the actual limits. - [What is the best way to save tokens in Codex?](https://halv.ai/blog/how-to-save-tokens-in-chatgpt/): Give Codex precise tasks, relevant files, and concise AGENTS.md instructions. Disable unused MCP servers. Measure tokens and cost separately. Halv's latest workflow benchmark used more tokens but recorded 57.1% lower model cost. ## Benchmark evidence - [SWE-rebench overview](https://halv.ai/blog/halv-57-percent-lower-model-cost/): Main results in plain language. - [Technical report](https://halv.ai/blog/halv-swe-rebench-astra-42-pairs/): Methodology, all 14 task outcomes, three repetitions, costs, and limitations. - [Evidence browser](https://halv.ai/evidence/swe-rebench-astra-42-pairs/): All 84 sanitized run summaries and recorded verifier rewards. - [Machine-readable results](https://halv.ai/evidence/swe-rebench-astra-42-pairs/selected-pairs.json): Scope, model configuration, aggregate metrics, and individual task results. - [Evidence verifier](https://halv.ai/evidence/swe-rebench-astra-42-pairs/verify-selected-pairs.mjs): Public script to check the published records and aggregate results. - [Evidence archive](https://halv.ai/evidence/swe-rebench-astra-42-pairs/halv-swe-rebench-astra-42-pairs-evidence.tar.gz): Downloadable summaries, session accounting, exclusions, checksums, and an arithmetic verifier. ## Optional - [Blog](https://halv.ai/blog/): Engineering notes, benchmarks, and usage guides. - [Privacy](https://halv.ai/privacy/): Desktop data handling, model requests, and website tracking. - [Terms](https://halv.ai/terms/): Software and model-provider terms. ## Coding agent usage guides - [AGENTS.md Token Usage: How Much Do Instructions Cost?](https://halv.ai/blog/agents-md-token-usage/): AGENTS.md instructions contribute context when Codex loads them. Their cost depends on the loaded text, tokenizer, caching, and subsequent work. Keep essential rules and remove duplication. Verify which files apply before moving instructions, because nested guidance depends on the discovery path. - [Claude Code vs Codex: Which Fits Your Workflow?](https://halv.ai/blog/claude-code-vs-codex/): Choose Claude Code or Codex by testing the tasks you repeat, the tools your project needs, and the review process you trust. Both support terminal workflows and cloud work. Neither has a universal quality or cost advantage. Keep the model, permissions, and billing basis visible when comparing results. - [Codex /compact vs /new: Which Saves More Tokens?](https://halv.ai/blog/codex-compact-vs-new/): Use /compact to continue the same task with a shorter conversation summary. Use /new when you need a fresh chat. Neither guarantees the lowest total usage: compaction requires work, while a new chat can require repeated investigation. Include that work and the final test result in your comparison. - [Halv is open source](https://halv.ai/blog/halv-is-open-source/): The Halv desktop app and its local savings engine are now open source under Apache 2.0, with no trial clock. Halv Cloud is the only paid part, at US$10 a month, because it runs on our servers and costs money to operate. - [How Much Context Do MCP Servers Add? A Token Usage Guide](https://halv.ai/blog/mcp-server-context-token-usage/): MCP has no fixed token cost per server. Usage depends on server instructions, tool definitions, the client's discovery behavior, and tool results. Some clients defer full definitions until needed. Measure the session and completed task before concluding that removing an MCP server saves money. - [How to Reduce ChatGPT Usage](https://halv.ai/blog/how-to-reduce-chatgpt-usage/): To reduce ChatGPT usage, keep one topic per chat and include only context that changes the answer. Limit unnecessary tools and request the exact output you need. Use Projects for reusable material. For software tasks, use Codex when you need repository tools and technical views. ChatGPT Work and Codex share usage, but ordinary Chat can follow different plan rules. - [How to Reduce Claude Usage](https://halv.ai/blog/how-to-reduce-claude-usage/): To reduce Claude usage, keep conversations focused, shorten Project instructions, remove unused files, disable unnecessary tools, and lower effort for routine work. In Claude Code, clear context between unrelated tasks and compact only for continuity. Claude, Claude Desktop, and Claude Code share plan usage, while API billing remains separate. - [How to Reduce Codex Usage](https://halv.ai/blog/how-to-reduce-codex-usage/): To reduce Codex usage, start a new chat for unrelated tasks, limit repository context, use only required tools, and match model effort to task difficulty. Track the complete workflow because workers, retries, cloud tasks, and reviews can consume the same plan allowance. Lower token use can extend an allowance, but it does not reduce a fixed subscription price. - [Inside Halv’s 57.1% cost reduction: the SWE-rebench data](https://halv.ai/blog/halv-swe-rebench-astra-42-pairs/): Halv recorded 57.1% lower model cost across 42 paired repetitions, with 25 verifier passes per arm. The strongest task result was ArcadeDB-4455: 81.7% lower cost and 3/3 passes on both sides. This interim workflow comparison used more tokens with Halv. Infrastructure-invalid attempts and a rolled-back policy experiment are excluded. - [Same pass count. 57.1% lower model cost.](https://halv.ai/blog/halv-57-percent-lower-model-cost/): In our latest SWE-rebench checkpoint, Halv recorded 57.1% lower model cost while matching vanilla Codex's 25 verifier passes. We tested 14 tasks three times each. This is a partial result for one workflow, not a promise for every task or a subscription discount. - [Why Is Codex Using So Many Tokens? 7 Things to Check](https://halv.ai/blog/why-is-codex-using-so-many-tokens/): Codex usage includes more than your prompt. Instructions, conversation history, file content, tool results, and model output can all contribute. Start with session usage, then check repeated work, reasoning effort, and subagents. Compare completed tasks, because a smaller request can still produce more retries.
| – |
| 6 Nov 2026 |
| – |
| 6 Nov 2026 |