Beyond one-shot answers. Into the codebase.

HHR means Hyper-Horizon Reasoning: a coding agent that can inspect a project, call tools, absorb the results, and keep going. It turns a high-level goal into concrete steps, previews edits, and checks the outcome. Models route through Kilo Code.

$ curl -fsSL https://hhr-ai.vercel.app/install.sh | bash
recommended curl — installs the full TUI · free models need no API key · Node.js 18+
~/dxn1 — hhr

What is hhr-ai?

HHR stands for Hyper-Horizon Reasoning: an iterative agent loop that reads your project, uses tools, incorporates what they return, and continues toward the goal. It shows its work and asks before edits or commands. Model requests route through Kilo Code.

Built around visible tool calls. Tool calls are visible, edits are reviewed, and shell commands require approval.

Next-token generation, streamed live. The selected model predicts text one token at a time. HHR streams the response as it arrives, so you can start reading before generation finishes.

No in-house models, ever. Prompts route through Kilo Code. hhr loads its live model catalog, marks free models, and uses your selected model for each response.

[*]
Kilo model catalog Search live models and see which ones are free
[*]
Next-token streaming Read responses as the model generates them for faster feedback
[*]
Free models Use Kilo's free models without adding an API key
[*]
Parallel subagents Delegate independent read-only investigations and combine findings
[*]
Any terminal Full-screen TUI, works wherever a shell runs
[*]
Closed core Installer is auditable; the reasoning core is proprietary

FAQ

What is hhr-ai?

HHR means Hyper-Horizon Reasoning. It is a terminal coding agent that can inspect code, use tools, absorb the results, and continue across multiple steps. You choose the model; prompts route through Kilo Code.

Which models does it use?

None of our own — and that's the point. hhr routes through Kilo Code and shows its live model catalog. Free models are marked in the model picker; paid models need a Kilo Code API key.

Does next-token prediction make HHR faster?

The selected model predicts text one token at a time, and HHR streams the response as it arrives. That gives you visible feedback before the full answer is complete; total generation speed depends on the selected model and network.

npm install hits "EACCES: permission denied"?

Your npm global folder is root-owned (Kali/Debian default). Fix it once, never use sudo again:

npm config set prefix ~/.npm-global — then relaunch your shell and npm i -g hhr-ai just works.

How do I install it?

One line: curl -fsSL https://hhr-ai.vercel.app/install.sh | bash. It downloads the full TUI straight from this website. Node.js 18+ is required; free Kilo models work without an API key. Also on npm: npm i -g hhr-ai@latest.

Do I need an AI subscription?

No subscription is needed to try Kilo's free models. Paid models may require a Kilo Code API key.

Can I use my existing subscriptions?

HHR currently routes all model requests through Kilo Code. Paid Kilo models require a Kilo Code API key.

How much does it cost?

The live catalog marks Kilo's free models. Paid models may incur charges through Kilo Code.

Is hhr-ai open source?

No — hhr-ai is deliberately closed-source. The installer is a single auditable script you can read before piping to bash, but the reasoning core and heuristic router remain proprietary.