InBoost LongRun Evaluation Runtime

Local context conservation and quota optimization runtime for AI coding assistants
Compatible with Claude Code, Google Antigravity (agy), Cursor, Windsurf, and OpenHands.

---

💡 Developer Advantages

Autonomous coding agents burn through token budgets and subscription rate limits rapidly when exploring codebases and debugging test suites. InBoost LongRun acts as a local intelligence filter, keeping your agent's context clean and focused.

• Extended Rate-Limit Runway: Work through longer, complex tasks without hitting hourly message caps or provider quota exhaustion walls.
• Lower Token Burn & Costs: Prevent agents from ingesting thousands of unnecessary tokens from full file dumps and massive test framework stack traces.
• Faster Agent Response Times: Leaner context windows significantly decrease time-to-first-token and overall turn latency.
• Zero Wasted Repair Loops: Local validation intercepts broken edits on your CPU before they are saved, saving 2–5 expensive roundtrips to fix trivial syntax errors.
• Complete Privacy Guarantee: 100% of processing happens on your local CPU. Your code and prompts never leave your workstation.

---

🚀 Quick Install

Option 1: One-Line Installers (Linux / macOS / Windows)

Linux & macOS:

curl -fsSL https://raw.githubusercontent.com/inboost-dev/inboost-ai-longrun-runtime/main/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/inboost-dev/inboost-ai-longrun-runtime/main/install.ps1 | iex

Option 2: Node.js (npm / npx)

# Run instantly with zero installation via npx:
npx inboost-longrun status
npx inboost-longrun serve

# Or install globally:
npm install -g inboost-longrun

Option 3: Python (pip)

pip install inboost-longrun
inboost-longrun status

Option 4: Local Repository Clone

git clone https://github.com/inboost-dev/inboost-ai-longrun-runtime.git
cd inboost-ai-longrun-runtime
./install.sh

---

🛠 Model Context Protocol (MCP) Tools

InBoost LongRun connects seamlessly to your editor or agent via the standard Model Context Protocol (MCP), providing high-speed local tools designed to maximize context efficiency:

Tool	Purpose & Developer Advantage
l0_view_outline	Structural Code Inspection: Gives your AI assistant a clear map of symbols, functions, and definitions without spending thousands of tokens reading full file bodies.
l0_clean_test_log	Noise-Free Error Isolation: Strips framework noise from test logs (pytest, unittest, jest) to expose the root-cause failure, preventing context pollution.
l0_verify_syntax	Instant Syntax Guardrail: Verifies code correctness on your local CPU before saving, preventing wasted prompt turns on syntax errors.
l0_verify_edit_scope	Targeted Change Guardrail: Keeps code modifications focused and within budget, stopping runaway diffs before they happen.
l0_run_test	Compact Test Execution: Runs local test suites and returns clean, signal-dense output tailored for agent decision-making.
l0_savings_report	Savings & Quota Ledger: Reports cumulative tokens conserved, quota windows extended, and syntax errors intercepted during your active session.

---

🤖 AI Agent Setup

1. Claude Code

claude mcp add inboost-longrun -- inboost-longrun serve

Add optimization rules to .claude/rules/inboost_longrun.md (or CLAUDE.md):

# InBoost LongRun Optimization Directives

1. Context Conservation: Before reading large files, use l0_view_outline to inspect declarations and structure.
2. Test Failure Isolation: Filter test logs and runtime exceptions via l0_clean_test_log to eliminate noise and isolate failing lines.
3. Syntax Integrity: Validate all code modifications with l0_verify_syntax prior to finalizing changes.
4. Edit Scope Compliance: Ensure edits remain focused and within budget. Verify with l0_verify_edit_scope.
5. Execution Optimization: Run test commands via l0_run_test for compact results. Check session economy with l0_savings_report.

2. Google Antigravity (agy)

agy mcp add inboost-longrun inboost-longrun serve

Rules are automatically loaded from .agents/rules/inboost_longrun.md.

3. Cursor

Add inboost-longrun to .cursor/mcp.json:

{
  "mcpServers": {
    "inboost-longrun": {
      "command": "inboost-longrun",
      "args": ["serve"]
    }
  }
}

---

📦 Supported Platforms

All release binaries are standalone, statically compiled or native executables with zero external dependencies:

• Linux: x86_64 & arm64 (Universal Static ELF, zero libc dependencies)
• macOS: Apple Silicon (arm64, M1–M4) & Intel (x86_64) (Native Mach-O)
• Windows: x86_64 (Native Console PE32+)

---

🔒 Privacy & Data Security

InBoost LongRun executes 100% locally on your machine and never transmits code externally.

All analysis, verification, and filtering routines run exclusively on your workstation's CPU. The engine makes zero network connections, sends no telemetry, and requires no external API keys.

---

🧪 Testing & Verification

Run the automated test suite to verify binary integrity, license validity, and MCP protocol compliance:

./tests/run_tests.sh

---

⚖️ Evaluation License Terms

Usage of this distribution is subject to the INBOOST LONGRUN EVALUATION LICENSE AGREEMENT:

• Evaluation Purpose Only: Licensed exclusively for evaluation and non-production testing.
• Term Limit: Access is limited to the active evaluation period. Use must cease upon expiration.
• Disclaimer: Provided "AS IS" with no warranties; all liability is disclaimed to the maximum extent permitted by law.
• Public Benchmarks: Publication of benchmark results requires prior written consent from InBoost.
• Jurisdiction: InBoost reserves the right to determine the governing law and dispute resolution jurisdiction.