Skip to content

RouteNex v0.2.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 21 Oct 12:07
· 4 commits to main since this release
v0.2.0
a5f0e68

RouteNex v0.2.0 — Official Open-Source Release Notes

RouteNex: The Intelligent Open-Source AI Gateway & Agent Control Plane.
One endpoint for every AI model. Deterministic routing, 3-layer resilience, 0ms caching, 4-stage LexiPack token optimization, and zero vendor lock-in.

🚀 Key Highlights in v0.2.0

1. Unified Standalone Architecture

  • Single Command Execution: npx routenex or docker compose up -d starts both the high-throughput Hono gateway (:3001) and the static web console.
  • Zero Cloud Dependency: Operates 100% offline with local AES-256-GCM keystores, SHA-256 virtual access keys, and local SQLite WAL persistence.
  • Open Source Gateway Auth Gate: Protected by default local setup password (routenex-admin), configurable via terminal or environment variables.

2. 3-Layer Resilience Engine

Survives upstream AI provider outages and rate limits without dropping in-flight agent workflows:

  • Layer 1 (Provider Circuit Breaker): Automatically trips on 3 consecutive 5xx failures or timeouts; initiates a 30-second cooldown with half-open recovery probes.
  • Layer 2 (Key Cooldown): Automatically parses Retry-After headers on HTTP 429 rate limits, cooling down only that specific provider key while sibling keys in the pool continue serving traffic.
  • Layer 3 (Model Lockout): If a specific upstream model returns HTTP 503 or 404, RouteNex isolates only that individual model ID without affecting sibling models under the same provider.

3. Free-Tier Pool Aggregator (~1.62 Billion Free Tokens/mo)

Connects directly to verified zero-cost and developer allowances across:

  • Mistral AI: 1 Billion tokens/month (codestral-latest, mistral-small-latest).
  • Pollinations AI: ~500M tokens/month (Zero-key anonymous).
  • Cerebras Cloud: 1,000,000 tokens/day at 2,000 tok/sec (cerebras/llama-3.3-70b).
  • Groq Cloud: 14,400 requests/day (groq/llama-3.3-70b-versatile, llama-3.1-8b).
  • SiliconFlow: ~50M tokens/month (DeepSeek-V3, DeepSeek-R1).
  • Google AI Studio: 1,500 requests/day (gemini-2.5-flash, gemini-2.5-pro).
  • DuckDuckGo AI: ~20M tokens/month (Zero-key anonymous).

4. 4-Stage LexiPack Token Optimizer

Compresses prompt payload size by 30% to 55% before transmitting over the wire, saving API costs and latency:

  • Stage 1 (Lite): Strips HTML comments, duplicate whitespace, and multi-line breaks.
  • Stage 2 (Log Pruner): Truncates repeated stack traces, terminal noise, and build output.
  • Stage 3 (Deduplication): Identifies and dedupes repetitive multi-turn conversation context.
  • Stage 4 (LexiPack Compact): Strips polite conversational fluff while preserving code blocks, markdown tables, JSON, and URLs byte-for-byte.

5. Multi-Model Fallback Combos & 19 Routing Strategies

  • Pre-configured clusters: combo/free-unlimited, combo/coding-power, combo/reasoning-trio.
  • Strategies: cost (floor price), latency (nitro speed), balanced, lkgp (last-known-good provider), headroom (failure rate weighted), reset-window, and cache-optimized.

6. One-Command Coding Agent Integrations

Spawns developer tools with RouteNex environment variables and isolated virtual keys injected per-process without touching global shell profiles:

npx routenex run claude --model auto
npx routenex run aider --model combo/coding-power
npx routenex run cursor
npx routenex run cline

7. Native Model Context Protocol (MCP) Server

Built-in JSON-RPC 2.0 MCP server exposing 5 autonomous tools for agent IDEs:

  • list_models: Filter by capabilities, free-only status, or provider.
  • explain_route: Dry-run routing evaluation detailing candidates and scores.
  • get_free_tier_quotas: Real-time tracking of developer allowances and failures.
  • test_provider_health: Active connectivity check.
  • upsert_combo: Dynamic fallback cluster creation.
  • Supports both Stdio MCP (Claude Desktop, Cursor) and Remote HTTP MCP (/v1/mcp).

8. SQLite WAL Persistence & Storage Maintenance

  • SQLite WAL mode (PRAGMA journal_mode = WAL) with busy timeout and bounded 5,000 LRU ring buffer.
  • Built-in /v1/admin/storage/vacuum maintenance and /v1/admin/storage/export JSON export.

📦 Quickstart

With npx:

npx routenex

Open http://127.0.0.1:3001/dashboard in your browser.

With Docker:

docker run -d -p 3001:3001 -v routenex-data:/app/data --name routenex harshil1876/routenex:latest

Python SDK Drop-in:

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:3001/v1",
    api_key="rn-local-dev-default"  # or optional on localhost
)

response = client.chat.completions.create(
    model="combo/free-unlimited",
    messages=[{"role": "user", "content": "Explain quantum teleportation"}]
)
print(response.choices[0].message.content)

TypeScript SDK Drop-in:

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'http://127.0.0.1:3001/v1',
  apiKey: 'rn-local-dev-default',
});

const response = await client.chat.completions.create({
  model: 'auto',
  messages: [{ role: 'user', content: 'Explain quantum teleportation' }],
});
console.log(response.choices[0].message.content);

🧪 Verification & Quality Metrics

  • Vitest Integration Tests: 73 tests passed (0 failures).
  • Static Build: 217 static pages compiled with zero errors.
  • Security Guardrails: OWASP security headers, SSRF private IP blocker, 10MB payload cap, timing-safe string comparison, brute-force rate limiter.
  • License: MIT License & Apache License 2.0 dual-license.

Full Changelog: https://github.com/harshil1876/RouteNex/commits/v0.2.0

Assets 2
Loading