README.md

---
title: HotCopy.ai
emoji: 🔥
colorFrom: red
colorTo: gray
sdk: static
pinned: true
short_description: Recursive Language Model CLI on Cloudflare Workers.
tags:
  - recursive-language-models
  - agentic-coding
  - cloudflare-workers
  - cli
  - rlm
  - arxiv:2512.24601
---

HotCopy is a Recursive Language Model CLI on Cloudflare Workers. Unbounded context, agent swarms in parallel, two-tier orchestrator/worker architecture (256K context per agent), no API keys, no config.

<div class="hc-org-card">
  <style>
    .hc-org-card {
      background: #0a0a0a;
      color: #e8e8e8;
      font-family: 'IBM Plex Mono', ui-monospace, SFMono-Regular, monospace;
      font-weight: 400;
      padding: 32px 24px;
      border-radius: 8px;
      margin: 16px 0 32px;
    }
    .hc-org-card .hc-stamp {
      display: inline-block;
      font-size: 12px;
      letter-spacing: 0.12em;
      color: #d4bf8a;
      border: 1px solid #d4bf8a;
      padding: 4px 8px;
      margin-bottom: 24px;
    }
    .hc-org-card .hc-wordmark {
      font-size: 30px;
      font-weight: 600;
      letter-spacing: -0.02em;
      margin: 0 0 16px;
      line-height: 1.1;
    }
    .hc-org-card .hc-wordmark .glyph {
      color: #e05030;
    }
    .hc-org-card .hc-tagline {
      font-size: 16px;
      line-height: 1.5;
      color: #e8e8e8;
      max-width: 56ch;
      margin: 0 0 24px;
    }
    .hc-org-card .hc-tagline em {
      font-style: normal;
      color: #d4bf8a;
    }
    .hc-org-card .hc-cta-row {
      display: flex;
      flex-wrap: wrap;
      gap: 16px;
      margin-top: 8px;
    }
    .hc-org-card .hc-cta {
      display: inline-block;
      font-size: 14px;
      font-weight: 600;
      letter-spacing: 0.04em;
      padding: 12px 24px;
      text-decoration: none;
      border: 1px solid #e05030;
      border-radius: 4px;
    }
    .hc-org-card .hc-cta-primary {
      background: #e05030;
      color: #0a0a0a;
    }
    .hc-org-card .hc-cta-secondary {
      background: transparent;
      color: #e05030;
    }
    .hc-org-card .hc-meta {
      font-size: 12px;
      color: #d4bf8a;
      letter-spacing: 0.08em;
      margin-top: 24px;
    }
  </style>
  <div class="hc-stamp">▸ OPS-BRIEF / SECTOR: DEV-INFRA</div>
  <h1 class="hc-wordmark"><span class="glyph">h.</span>HotCopy.ai</h1>
  <p class="hc-tagline">Your codebase is too big for any context window. <em>So we killed the context window.</em> Managed recursive AI coding CLI — agent swarms decompose impossible problems in parallel.</p>
  <div class="hc-cta-row">
    <a class="hc-cta hc-cta-primary" href="https://hotcopy.ai">Install</a>
  </div>
  <div class="hc-meta">▸ UNBOUNDED CONTEXT // 50+ PARALLEL AGENTS // 0 API KEYS</div>
</div>

## What HotCopy is

Managed recursive AI coding CLI. Agent swarms decompose impossible problems in parallel. No context limits. No API keys. No config. Just install and go. HotCopy treats your project as an *environment* the model interacts with programmatically — never raw files in a prompt.

## The Recursive Language Model paradigm

HotCopy implements the Recursive Language Model (RLM) paradigm introduced by Zhang, Kraska, and Khattab at MIT CSAIL ([arXiv:2512.24601](https://arxiv.org/abs/2512.24601), Jan 2026). Existing scaffolds (Claude Code, Gemini CLI, Cursor) place the user prompt directly into the LLM's context window, generate output autoregressively (capped by token limits), and verbalize sub-LLM calls (only O(1) delegations). RLMs invert this: the prompt is loaded as a REPL variable, the model writes code to interrogate it, sub-calls are programmatic functions inside that code, and the answer is built up in variables and returned via `FINAL_VAR()` — unbounded by any single model's generation length.

## Benchmarks vs base model

| Benchmark | Input Size | Base GPT-5 | RLM(GPT-5) | Avg Cost |
|---|---|---|---|---|
| BrowseComp+ | 6–11M tokens | **0%** | **91.3%** | $0.99 |
| OOLONG | 131K tokens | 44% | 56.5% | $0.43 |
| OOLONG-Pairs | 32K tokens | 0.04% | 58.0% | $0.33 |
| CodeQA | 23K–4.2M tokens | 24% | 62% | $0.11 |

Median RLM cost ≤ base model cost; up to 3× cheaper than summary-agent baselines. Source: Zhang, Kraska, Khattab. *Recursive Language Models.* MIT CSAIL. [arXiv:2512.24601](https://arxiv.org/abs/2512.24601), Jan 2026.

## How it works

- **Prompt as environment.** Project context is loaded into Durable Object memory as REPL variables. The orchestrator sees only metadata (file count, repo map, total chars) — never raw files.
- **Two-tier orchestrator/worker architecture.** The RLM root model owns a 256K orchestration window. Every sub-call routes to a separate worker model with its own 256K window and multimodal capability — managed for you, no provider keys to wire up.
- **REPL with stdout truncation.** The orchestrator writes JavaScript that runs in a sandboxed V8 isolate (`@cloudflare/codemode`). Stdout is truncated before being appended to history — variables hold the long state, the model history holds the metadata.
- **Parallel sub-calls via `llm_batch()`.** Unlike the paper's sequential implementation, HotCopy fans out scouts in parallel through Cloudflare Workers, dramatically reducing latency on map-reduce decompositions.
- **Durable Objects + AI Gateway.** Every model call routes through `hotcopy-gateway` for cost tracking, prefix caching (per-session affinity), and rate-limit handling. State lives in DO SQL — trajectories, sub-calls, costs, all logged for debugging.
- **`FINAL_VAR()` for unbounded output.** The model returns the *name* of the variable holding the answer. The reply is whatever that variable holds — no autoregressive output cap.

## Architecture deep-dive

<details>
<summary><strong>REPL scope, three-tier memory, and the permissions model</strong></summary>

### REPL scope (root code)

| Category | Functions |
|---|---|
| Context | `context.read(path)`, `context.readAll(paths)`, `context.search(pattern)`, `context.preview(chars)`, `context.metadata`, `context.files` |
| Sub-calls | `llm_query(prompt)`, `llm_batch(prompts)` |
| Agents | `spawn_worker(role, task, ctx)`, `vision_analyze(base64, prompt)` |
| Memory | `memory_search(query, limit)`, `memory_failures(query)`, `memory_save(type, content)` |
| File tools | `write_file(path, content)`, `edit_file(path, old, new)`, `shell(cmd, cwd)` |
| Verification | `verify(claim, evidence)`, `verify_claims(...)` |
| Persistent state | `set_context(label, content)`, `load_context(label)` |
| Terminal | `FINAL(text)`, `FINAL_VAR(varName)`, `print(...)` |

### Three-tier memory

- **Working memory** — per-agent context window. Compaction at 70–75% utilization. Budget: 12% system, 18% tools, 25% retrieved context, 25% history, 5% user input, 15% output reserve.
- **Session memory** — `SWARM_STATE.md` task manifest, continuously updated by the orchestrator. Exploits transformer recency bias by placing the live task list at the end of context.
- **Project memory** — D1 + Vectorize. Observations, decisions, failures indexed for semantic search. AI-compressed after N turns. Exposed via MCP for external tool access.

### Permissions

Default-deny on `fetch_*`, `browse`, `crawl`, `shell`. Wildcard rules in D1 (`permission_rules`) let users selectively allow patterns (`api.github.com/*`, `git status`, etc.) at three persistence tiers — session, user, or global. Every quarantined call hits a tier-cascade evaluator; first match wins; "ask" rules round-trip to the CLI for a per-call decision.

### Stack

Cloudflare Workers + Agents SDK · Durable Objects (RLMOrchestrator, WorkerAgent, VisionAgent, MemoryAgent) · D1 · Vectorize · R2 · Dynamic Workers (`@cloudflare/codemode`) · AI Gateway (`hotcopy-gateway`).

</details>

## Install

```bash
npm i -g hotcopy && hotcopy
```

Sign in on first run. No API keys. No config files. Managed inference is included.

## What we publish on Hugging Face

<div class="hf-grid">
  <style>
    .hc-org-card + .hf-grid, .hf-grid {
      display: grid;
      grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
      gap: 16px;
      margin: 24px 0;
    }
    .hf-grid .hf-card {
      background: #0a0a0a;
      color: #e8e8e8;
      border: 1px solid #d4bf8a;
      border-radius: 4px;
      padding: 16px;
      text-decoration: none;
      font-family: 'IBM Plex Mono', ui-monospace, SFMono-Regular, monospace;
      font-weight: 400;
    }
    .hf-grid .hf-card:hover {
      border-color: #e05030;
    }
    .hf-grid .hf-card .hf-emoji {
      font-size: 20px;
      display: block;
      margin-bottom: 8px;
    }
    .hf-grid .hf-card .hf-title {
      font-size: 14px;
      font-weight: 600;
      color: #e05030;
      display: block;
      margin-bottom: 4px;
    }
    .hf-grid .hf-card .hf-sub {
      font-size: 12px;
      color: #d4bf8a;
      display: block;
    }
  </style>
  <a class="hf-card" href="https://huggingface.co/spaces/HotCopyAI/hotcopy-rlm-demo">
    <span class="hf-emoji">🚀</span>
    <span class="hf-title">Showcase Space</span>
    <span class="hf-sub">Live RLM demo</span>
  </a>
  <a class="hf-card" href="https://huggingface.co/datasets/HotCopyAI/rlm-trajectories-seed">
    <span class="hf-emoji">📊</span>
    <span class="hf-title">RLM Trajectories</span>
    <span class="hf-sub">Open dataset</span>
  </a>
  <a class="hf-card" href="https://hotcopy.ai">
    <span class="hf-emoji">🌐</span>
    <span class="hf-title">hotcopy.ai</span>
    <span class="hf-sub">Product site</span>
  </a>
  <a class="hf-card" href="https://arxiv.org/abs/2512.24601">
    <span class="hf-emoji">📄</span>
    <span class="hf-title">Paper</span>
    <span class="hf-sub">arXiv:2512.24601</span>
  </a>
</div>

---

Contact: hello@hotcopy.ai · [hotcopy.ai](https://hotcopy.ai)