SO WE KILLED
THE CONTEXT
WINDOW.
Managed recursive AI coding CLI. Agent swarms decompose impossible problems in parallel.
No context limits. No API keys. No config. Just install and go.
Recursive Language
Model architecture.
A root orchestrator decomposes your task. Worker agents execute in parallel. Results synthesize back. No single model sees everything — but everything gets seen.
One npx command. Log in with GitHub. We handle models, orchestration, scaling, routing.
Orchestrator dispatches worker agents. Each gets targeted context. Simultaneous execution across your project.
Orchestrator validates, applies changes. If something's off, it recurses — corrective agents until clean.
Fully managed.
You just write code.
We run the models, orchestration, routing, and scaling on hardened infrastructure. Your source stays local. Only targeted context reaches agents in encrypted, ephemeral channels. Code is never stored or trained on. Decisions and patterns persist in your private memory vault.
50+ agents.
One command.
One bill.
We run the models, orchestration, routing, and scaling on hardened infrastructure. Your source stays local. Only targeted context reaches agents in encrypted, ephemeral channels. Code is never stored or trained on.
For the codebase
nobody fully
understands anymore.
Not the 50-file demo project. The 2-million-line monorepo with legacy PHP mixed with modern TypeScript.
Parallel agent swarms
10, 20, 50+ worker agents analyzing different parts of your codebase simultaneously. Security audit, test generation, and docs happening in parallel. Wall-clock time drops 5-10x.
One command to migrate
hotcopy import --from claude-code brings your CLAUDE.md, MCP servers, permissions, and hooks. Same for Gemini CLI. Existing setup transfers in seconds.
No API keys. No cloud setup.
Authenticate with GitHub and go. We manage the models, routing, scaling, and infrastructure. You manage your code. That's the deal.
Scoped. Ephemeral. Never stored.
Only files relevant to each task reach agents in scoped, encrypted chunks. Source is never persisted or trained on. Only agent memories — decisions, patterns, failure records — persist to make future sessions smarter.
Predictable pricing
Generous included tasks on every plan. No per-token billing, no counting input vs output, no surprise invoices. Overage is transparent and billed at the same rate. Ship code, not spreadsheets.
Recursive context decomposition
No agent sees your whole codebase. Each gets surgically targeted context — relevant files, dependency maps, task scope. The system sees everything through decomposition.
Agents that remember.
Across sessions.
Across teams.
Other coding tools forget everything when the session ends. Hotcopy's Blackbox memory engine gives your agents persistent intelligence that compounds over time.
Semantic memory search
Agents search past decisions, failures, and patterns using vector similarity. Before attempting any approach, they check what's been tried before.
Failure analysis
Every failed approach is categorized and stored. Agents query past failures before attempting fixes — they never repeat the same mistake twice.
DAG-based task planning
Complex tasks decompose into dependency-validated execution plans with parallel execution, cycle detection, and heartbeat tracking.
Session handoffs
Stop work on your laptop, pick it up on your desktop. Cross-machine transfer tokens carry full session context — nothing is lost.
Anti-pattern detection
Detects when agents are looping, yak-shaving, or going off-track. Convergence tracking breaks unproductive cycles before they burn tokens.
Shared team memory
When one developer's agents discover a rate limit gotcha or API quirk, every team member's agents know it immediately. Knowledge compounds across the org.
Simple plans.
Predictable bills.
Every paid plan includes tasks at a discount. Overage is $1.30/task — still 5× cheaper than the competition.
Kick the tires. See what recursive agent swarms can do on your real codebase.
- 10 tasks / month
- Up to 3 parallel agents
- Standard routing
- Community support
- CLI access
For developers who ship daily. Full agent power, priority routing, and $130 in included tasks.
- 100 tasks included / month
- Up to 12 parallel agents
- Priority routing
- Blackbox persistent memory
- HOTCOPY.md custom config
- MCP server support
- Import from Claude Code / Gemini
- $1.30 / task overage
For teams running full-scale operations. $650 in included tasks per seat — the more you use, the more you save.
- 500 tasks / seat / month
- Up to 50+ parallel agents
- Dedicated model capacity
- Blackbox team shared memory
- SSO & audit logs
- SLA & priority support
- Custom routing rules
- Volume discounts on overage
Same task.
5× cheaper.
We ran a real security audit — 52 files, 82 parallel scouts, 15 orchestrator turns. Here's what you'd pay.
▸ 82 agents // parallel // MINUTES
▸ sequential // tool calls // HOURS
Full security audit of a 52-file codebase. 15 orchestrator turns. 82 parallel scout agents analyzing files for vulnerabilities, generating fixes, and validating results.
What Hotcopy does
that others can't.
Parallel agents and MCP are table stakes. Here's where Hotcopy actually pulls ahead.
| ▸ CAPABILITY | ▸ HOTCOPY | ▸ CLAUDE CODE | ▸ CURSOR |
|---|---|---|---|
| Recursive decomposition | ✓ RLM engine | ✕ | ✕ |
| Context rot | ✓ Eliminated — scoped chunks | 1M window, degrades | 200K, degrades |
| Persistent memory | ✓ Blackbox engine | File-based only | ✕ |
| Learns from failures | ✓ Checks before retrying | ✕ | ✕ |
| Team shared memory | ✓ Cross-org | ✕ | ✕ |
| 52-file security audit | $1.30 | ~$6.68 (Opus) | Varies (opaque) |
| Predictable pricing | ✓ Included + overage | Opaque caps | $200-$1,800+/mo |
| Import config | ✓ One command | ✕ | ✕ |
| Parallel agents | ✓ 50+ scouts | Subagents | Up to 20 |
| MCP support | ✓ | ✓ | ✓ |
Roger that.
Let's deploy.
Private beta is filling up. Drop your email to get early access.