Skip to content
NEW BENCHMARK · SWE-REBENCH20 paired tasks

The desktop workspace for AI coding agents.

51.1%fewer tokens per correct answerSWE-rebench · 20 paired tasks

Codex + Halv4.43MCodex9.06M
Tokens per correct answercompared with Codex alone

Run Claude Code, Codex, Kimi, and GLM side by side in one desktop app. Keep your subscriptions. Halv cuts wasted context while you work.

macOS · Windows · Linux
10/20

Tasks solvedCodex: 7/20

30.2%

fewer tokens overall

46.9%

lower cost per correct answer

Measured on the first 20 paired SWE-rebench tasks with Codex. Results vary by task and model.

Your agents. One place to build.

Open a project. Choose your agent. Work in chat or native terminals, with your sessions, changes, and savings in one place.

Halv desktop workspace with Claude Code and Codex in three terminal panes, project history, and a savings meter.Halv desktop workspace with Claude Code and Codex in three terminal panes, project history, and a savings meter.
Inside the Halv desktop app

Bring your favorite agents.

Use Claude Code, Codex, Kimi, and GLM with the subscriptions you already have. Run multiple agents together.

Make room for real work.

Move between chat and native terminals. Split panes across tasks, organize projects, and return to saved sessions.

Keep control of the work.

Review changes, set approvals, and track agent activity and token savings from your workspace.

02 / THE PROOFSWE-rebench

Less budget. More working code.

We ran the same 20 repository tasks with and without Halv. The repository verifiers accepted 10 Halv answers and 7 vanilla answers.

20 TASKS. EVERY OUTCOME.
✓ Verifier passed– Verifier did not pass

Count the work that passes.

Token totals include all runs, including failed attempts. Divide by verified answers to see the budget each correct answer took.

METRICCodexCodex + HalvHALV CHANGE
Tasks solved7 / 2010 / 20+3 tasks
Total tokens63,438,10944,289,991-30.2%
Recorded cost$1.84932$1.40301-24.1%
Tokens per correct answer9,062,5874,428,999-51.1%
Cost per correct answer$0.26419$0.14030-46.9%

Same tasks. Same model. Same verifier.

gpt-5.6-luna · medium reasoning · priority service · Codex 0.152.0

The Halv arm used Crux, RTK, and Halv Engine Headroom together. This benchmark measures the complete toolkit.

An early result from 20 task pairs, published by Halv. Halv used fewer tokens on 12 tasks and more on 8. This is not a guarantee for every run.

03 / THE HALV ENGINE

Less waste, built into the workspace.

Halv helps your agents find the right code, trims noisy command output, and compresses model context as you work.

Headroom

CONTEXT COMPRESSION

Duplicated files, stale history, repeated instructions — removed. What changes the answer stays. When a request can't be compressed safely, it passes through untouched.

Crux

CODE NAVIGATION

Callers, references, impact — answered from an index instead of reading the repo a file at a time. Agents ask the index and stop burning context to orient themselves.

RTK

COMMAND OUTPUT FILTERING

Test logs, diffs, dependency trees and install chatter are trimmed to the part that matters before a single token is spent on them.

04PRICING

Big savings. Small price.

Halv is built to save you far more than it costs. Get more work from the agent subscriptions you already pay for.

Basic

$2/ MONTH

Every feature, with a monthly savings cap. Start small and see how much wasted context Halv can cut from your daily work.

Try it free. Let your savings make the case.

Track saved tokens in the app before you choose a plan. Your agent subscription price stays the same; Halv helps you get more work from it.

INCLUDED ON BOTH PLANS

  • Pro mode terminals
  • Four agent CLIs
  • Up to 32 live panes
  • Ghostty on macOS
  • Subagent fan-out
  • Crux code index
  • Command filtering
  • Live savings meter
  • Local OS sandbox
  • Per-project approvals
  • Turn diff & revert
  • macOS · Windows · Linux
THE ONLY DIFFERENCEAll features on both plans. Basic caps monthly savings; Unlimited removes that cap.

FREE TO TRY · CANCEL IN TWO CLICKS · WORKS WITH YOUR CLAUDE, CHATGPT, MOONSHOT OR Z.AI PLAN

05FAQ

Questions we get asked.

  • What is Halv?

    Halv is a 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. Halv combines code navigation, command output filtering, and context compression to reduce wasted tokens while you work.

  • How can I increase my Claude Code usage limit?

    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?

    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?

    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?

    Give Codex precise tasks, relevant files, and concise AGENTS.md instructions. Disable MCP servers you do not need. Halv combines code indexing, command output filtering, and context compression to reduce wasted tokens. In Halv's first 20 paired SWE-rebench tasks, Codex with Halv used 30.2% fewer total tokens and 51.1% fewer tokens per correct answer. This small, self-run benchmark does not guarantee the same savings on every task.

  • Does Halv work with Codex and Claude Code?

    Yes. Halv supports Claude Code, Codex, Kimi, and GLM. In Pro mode, each agent runs through its native CLI inside Halv. Sign in with your existing agent account and use the subscriptions you already pay for. API keys are optional for providers that support them.

  • How does Halv reduce coding agent token usage?

    Halv reduces token usage in three ways. Crux gives agents a code index for navigation and references. RTK filters noisy command output before it enters model context. Halv Engine Headroom compresses model context locally. Together, these tools reduce repeated code exploration and unnecessary context.

  • What evidence supports Halv's token savings?

    In Halv's first 20 paired SWE-rebench tasks, Codex with Halv used 51.1% fewer tokens per correct answer and 30.2% fewer total tokens. Repository verifiers accepted 10 of 20 Halv answers versus 7 of 20 Codex answers. Both arms used the same model, tasks, and verifier. Halv published this small, self-run benchmark and all 40 selected run records.

  • Will every task use 51.1% fewer tokens with Halv?

    Savings vary by task, model, and context. The 51.1% result measures total tokens divided by verifier-passing tasks, including tokens from unsuccessful attempts. Halv used fewer tokens on 12 benchmark tasks and more on 8. This Codex benchmark does not establish the same savings for Claude Code or guarantee identical answers.

  • How do Halv's token savings affect my subscription?

    Halv helps you get more work from your existing agent usage budget. Your provider's fixed subscription price stays the same. Lower token usage can reduce metered API costs, depending on pricing and caching. Compare your measured savings with Halv's price to see whether it pays for your workflow.

  • How is Halv different from a terminal or an MCP server?

    Halv combines coding agents, chat, native terminals, project history, approvals, and savings tracking in one desktop app. Crux is the code-index MCP server within Halv's toolkit. With a standalone CLI, you manage that workspace and those tools yourself. Halv brings them together around the agents you already use.

  • How much does Halv cost, and can I try it free?

    Basic costs US$2 per month with a monthly savings cap. Unlimited costs US$10 per month with no savings cap. Both plans include all features and a 7-day free trial. A card is required at signup. Cancel during the trial to pay nothing; otherwise, your subscription renews automatically.

  • Does Halv process my code locally?

    Halv runs code indexing and context compression on your computer. Your chosen model provider receives the prompts and context needed to answer your requests. Halv also connects to services for account access, billing, updates, and basic app events. Review agent actions and approvals before commands run or files change.

  • Does Halv run on macOS, Windows, and Linux?

    Yes. Halv supports macOS on Apple silicon and Intel, Windows x64, and Linux x64. Linux downloads include AppImage and deb packages. Halv is a desktop app; download and install it on a supported computer to use the workspace.

Less spend. More solved.

FROM ZERO TO SAVING

01

Download

macOS, Windows or Linux. No API key required.

02

Sign in to your agents

The CLIs you already use, on the plans you already pay for. Logins stay in the CLI.

03

Watch the meter

Savings start on your first prompt, receipt by receipt.

macOS · APPLE SILICON + INTEL

WINDOWS · x64

LINUX · AppImage + deb

SIGN IN AND TRY IT FREE.