Skip to content

okba14/NeuroHTTP

BranchesTags

Repository files navigation

⚡ NeuroHTTP — AI-Native Web Server (C + ASM)

High-performance, AI-native server built from scratch in C + hand-written Assembly (SSE2/AVX2/AVX-512).
Supports 10 AI providers — OpenAI, Groq, Anthropic Claude, Google Gemini, DeepSeek, Moonshot/Kimi, Zhipu GLM, Perplexity, Mistral AI, and local models via Ollama — all in one binary.

📖 Arabic Version — النسخة العربية

📋 Table of Contents

🔥 Features

⚡ Performance Core

  • Event-driven engine — single-threaded epoll edge-triggered event loop with io_uring detection
  • io_uring async I/O — full io_uring integration for zero-syscall I/O; falls back to epoll gracefully; SQPOLL support (Linux 5.11+)
  • Hand-written ASM — CRC32 (SSE4.2), JSON tokenizer (SSE2/AVX2), memory copy (SSE2–AVX-512), non-temporal stores
  • Zero-copy I/O — sendfile/splice support for efficient static file serving
  • Memory arena allocator — bump allocation + slab allocator for lock-free per-request memory
  • HTTP state-machine parser — no sscanf/regex, pure finite-state-machine HTTP/1.1 parser

🧠 Multi-Provider AI Routing

  • 10 providers — OpenAI, Groq, Anthropic, Gemini, DeepSeek, Moonshot/Kimi, Zhipu GLM, Perplexity, Mistral AI, Ollama
  • Smart routing — latency-aware, health-check-aware, automatic fallback on failure
  • Provider-agnostic format — models defined in config/aionic.conf via a simple pipe-delimited format

🛡️ Enterprise Security

  • Web Application Firewall — SQLi, XSS, path traversal detection; IP blacklist/whitelist
  • Token-bucket rate limiter — per-IP rate limiting with configurable RPS
  • Request inspection — suspicious user-agent, payload size limits
  • Content Moderation — built-in prompt/response moderation with regex-based filtering (hate speech, harassment, violence, self-harm, PII, malicious code)
  • Security Headers — CSP, X-Content-Type-Options, X-Frame-Options on all responses
  • ASLR + RELRO + Stack Protector — full binary hardening via PIE, _FORTIFY_SOURCE=2, -fstack-protector-strong
  • JSON Injection Protection — all user prompts properly escaped before insertion into API payloads

📊 Observability

  • Prometheus metrics — /metrics endpoint for scraping
  • Request tracing — per-request IDs, latency tracking
  • Provider stats — per-provider call count, latency, token usage, error rate
  • Server stats — /stats endpoint with live metrics

🧠 AI Gateway Features

  • Function Calling — parse and forward tools definitions to OpenAI-compatible providers; detect tool_calls in responses
  • Vision / Multi-modal — support for image inputs (image_url) in chat completions for GPT-4o, Claude 3.5, Gemini 2.0
  • Structured Outputs — response_format parameter support with JSON schema validation
  • Embeddings — /v1/embeddings endpoint for generating vector embeddings
  • Cost-Aware Routing — routing to cheapest available provider based on real-time per-model cost tracking; cost factor incorporated into model scoring (20% weight)

💰 Distributed Caching

  • In-memory cache — fast local cache with semantic similarity matching (n-gram based)
  • Redis backend — optional Redis distributed caching via cache_init_redis(); automatic fallback to in-memory if Redis unavailable
  • Semantic cache — fuzzy matching of cached responses for similar prompts

🔌 Extensibility

  • Plugin system — 6 hook points (pre-request, post-request, AI prompt, AI response, connect, disconnect)
  • GitHub Plugin Install — install plugins directly from GitHub releases via plugin_install_from_github("owner/repo", "*.so")
  • URL Plugin Install — install plugins from any URL via plugin_install_from_url()
  • Dynamic request buffer — no fixed-size limits on request bodies

🌐 Networking

  • TLS 1.3 — full TLS 1.3 support via OpenSSL 3.6 (or BoringSSL); OCSP stapling, ALPN negotiation, HSTS headers
  • HTTP/2 — h2 over TLS (ALPN) + h2c cleartext upgrade via nghttp2; multiplexed streams, server push, flow control
  • HTTP/1.1 keep-alive — connection reuse with configurable timeout
  • Server-Sent Events (SSE) — streaming endpoint at /v1/chat/stream
  • TCP defer accept — reduced accept overhead
  • Configurable SSL verification

🛡️ Graceful Shutdown

  • Connection draining — on SIGTERM/SIGINT, stops accepting new connections, drains active connections (with configurable timeout), then cleans up resources
  • Signal handling — SIGUSR1 triggers OCSP refresh, SIGHUP reloads API keys

🧩 Supported Providers

Provider Models Endpoint
Groq 🆓 Llama 3.3 70B, Llama 3.1 8B, Gemma 2 9B, Mixtral 8×7B, DeepSeek R1, Qwen QWQ 32B api.groq.com
OpenAI GPT-4o, GPT-4o-mini, GPT-4 Turbo, o3-mini api.openai.com
Anthropic Claude 3.5 Sonnet, Claude 3.5 Haiku, Claude 3 Opus api.anthropic.com
Google Gemini Gemini 2.0 Flash, Gemini 1.5 Pro generativelanguage.googleapis.com
DeepSeek DeepSeek Chat, DeepSeek Reasoner api.deepseek.com
Moonshot/Kimi Moonshot v1 8K, Moonshot v1 32K api.moonshot.cn
Zhipu (GLM) GLM-4-Plus, GLM-4-Flash open.bigmodel.cn
Perplexity Sonar Pro, Sonar Deep Research api.perplexity.ai
Mistral AI Mistral Large, Mistral Small api.mistral.ai
Local Ollama 🆓 Any local model via Ollama localhost:11434

Prerequisites

# Ubuntu / Debian
sudo apt-get update
sudo apt-get install -y build-essential nasm libcurl4-openssl-dev libssl-dev libnghttp2-dev liburing-dev

# Fedora / RHEL
sudo dnf install -y gcc nasm libcurl-devel openssl-devel libnghttp2-devel liburing-devel

# Arch Linux
sudo pacman -S --noconfirm gcc nasm libcurl-compat openssl nghttp2 liburing

# Verify NASM is installed (required for assembly files)
nasm --version   # Should output NASM version 2.x+

🛠 Build & Run

Build Commands

git clone https://github.com/okba14/NeuroHTTP.git
cd NeuroHTTP

# Production build (optimized with LTO, -O3, -march=native)
make

# Debug build (AddressSanitizer, UBSan, stack protector)
make debug

# Clean rebuild
make rebuild

# Build only plugins
make plugins

Run Commands

# Start the server (default port 8080)
./bin/aionic

# Run debug build
./bin/aionic-debug

# Or via Make
make run
make run-debug

The server will display:

========================================
    AIONIC AI Web Server v2.0.0
========================================
Build: May 29 2026
Features: TLS1.3 HTTP/2 io_uring OCSP
========================================
    - Port: 8080
    - TLS Port: 8443
    - Threads: 4
    - Max Connections: 1024
    - io_uring: enabled
    - Zero-Copy: enabled
    - TLS: disabled
    - HTTP/2: enabled
    - Smart Routing: enabled
    - Streaming: enabled
    - Graceful Shutdown Timeout: 30s
========================================

---

## ⚙️ Configuration

All configuration lives in `config/aionic.conf`. Here is every option explained:

### Server Settings

```ini
port = 8080                    # HTTP listen port
thread_count = 4               # Worker thread count (legacy, event loop uses main thread)
worker_threads = 4             # Event loop worker pool size
max_connections = 1024         # Maximum concurrent connections
request_timeout = 30000        # Request timeout in milliseconds (30s)
buffer_size = 8192             # Internal I/O buffer size
max_request_size = 33554432    # Maximum request body size (32 MB)
log_file = logs/aionic.log     # Log file path

Cache Settings

enable_cache = 1               # Enable response caching
cache_size = 1000              # Maximum cache entries
cache_ttl = 3600               # Cache TTL in seconds (1 hour)

Security Settings

enable_firewall = 1            # Enable WAF (SQLi, XSS, path traversal detection)
verify_ssl = 1                 # Verify SSL certificates for upstream AI endpoints
api_key = your-secret-api-key-here  # Server-wide API key for client authentication

TLS / HTTPS Settings (v2.0.0)

enable_tls = 0                      # Enable TLS 1.3 HTTPS (requires cert and key)
tls_port = 8443                     # HTTPS listen port
tls_cert_file = config/cert.pem     # TLS certificate path (PEM)
tls_key_file = config/key.pem       # TLS private key path (PEM)
tls_ca_file = config/ca.pem         # CA bundle for OCSP (optional)
tls_enable_ocsp = 0                 # Enable OCSP stapling
tls_ocsp_refresh_interval = 3600    # OCSP refresh interval in seconds
tls_hsts_max_age = 31536000         # HSTS max-age in seconds (1 year)

HTTP/2 Settings (v2.0.0)

enable_http2 = 1                    # Enable HTTP/2 (h2 via TLS ALPN + h2c upgrade)
http2_max_concurrent_streams = 256  # Max concurrent streams per connection
http2_max_header_list_size = 65536  # Max header list size
http2_initial_window_size = 65535   # Initial flow-control window size
http2_max_frame_size = 16384        # Max frame size

Engine Settings

enable_iouring = 1             # Enable io_uring async I/O (falls back to epoll)
enable_zero_copy = 1           # Enable sendfile/splice zero-copy I/O
enable_ratelimiter = 1         # Enable per-IP rate limiting
rate_limit_rps = 100           # Max requests per second per IP
enable_observability = 1       # Enable metrics and tracing
enable_streaming = 1           # Enable SSE streaming endpoint
enable_smart_routing = 1       # Enable latency/health-aware provider routing
enable_keepalive = 1           # Enable HTTP/1.1 keep-alive
keepalive_timeout = 30         # Keep-alive timeout in seconds

🔑 Adding API Keys

API keys are read from environment variables for security. Never hardcode keys in config files.

Setting API Keys

# Groq (free, get key at https://console.groq.com/keys)
export GROQ_API_KEY="gsk_your_groq_key_here"

# OpenAI (get key at https://platform.openai.com/api-keys)
export OPENAI_API_KEY="sk-projXXXXXXX"

# Anthropic Claude (get key at https://console.anthropic.com/)
export ANTHROPIC_API_KEY="sk-ant-your-anthropic-key"

# Google Gemini (get key at https://aistudio.google.com/apikey)
export GEMINI_API_KEY="AIza_your_gemini_key"

# DeepSeek (get key at https://platform.deepseek.com/)
export DEEPSEEK_API_KEY="sk_your_deepseek_key"

# Moonshot / Kimi
export MOONSHOT_API_KEY="your_moonshot_key"

# Zhipu GLM
export ZHIPU_API_KEY="your_zhipu_key"

# Perplexity
export PERPLEXITY_API_KEY="pplx_your_perplexity_key"

# Mistral AI
export MISTRAL_API_KEY="your_mistral_key"

Persisting Keys

Add them to your shell profile (~/.bashrc, ~/.zshrc, etc.):

echo 'export GROQ_API_KEY="gsk_your_key"' >> ~/.bashrc
echo 'export OPENAI_API_KEY="sk_your_key"' >> ~/.bashrc
source ~/.bashrc

Or use a .env file:

# .env
GROQ_API_KEY=gsk_xxxx
OPENAI_API_KEY=sk-xxxx

# Load before running
export $(grep -v '^#' .env | xargs) && ./bin/aionic

Adding AI Models

Models are defined in config/aionic.conf using this format:

ai_model = <name>|<endpoint>|<env_var>|<max_tokens>|<temperature>
Field Description Example
name Model identifier (used in API calls) gpt-4o
endpoint Full API URL https://api.openai.com/v1/chat/completions
env_var Environment variable name for the API key OPENAI_API_KEY
max_tokens Maximum response tokens 16384
temperature Sampling temperature (0.0–1.0) 0.7

Examples

# OpenAI-compatible (OpenAI, Groq, Together, vLLM, LocalAI, Ollama)
ai_model = gpt-4o|https://api.openai.com/v1/chat/completions|OPENAI_API_KEY|16384|0.7
ai_model = llama-3.3-70b-versatile|https://api.groq.com/openai/v1/chat/completions|GROQ_API_KEY|8192|0.7

# Anthropic Claude (uses /v1/messages endpoint)
ai_model = claude-3-5-sonnet-20241022|https://api.anthropic.com/v1/messages|ANTHROPIC_API_KEY|8192|0.7

# Google Gemini (uses generateContent endpoint)
ai_model = gemini-2.0-flash|https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent|GEMINI_API_KEY|8192|0.7

# DeepSeek
ai_model = deepseek-chat|https://api.deepseek.com/v1/chat/completions|DEEPSEEK_API_KEY|8192|0.7

# Moonshot / Kimi
ai_model = moonshot-v1-8k|https://api.moonshot.cn/v1/chat/completions|MOONSHOT_API_KEY|8192|0.7

# Zhipu GLM
ai_model = glm-4-plus|https://open.bigmodel.cn/api/paas/v4/chat/completions|ZHIPU_API_KEY|8192|0.7

# Perplexity
ai_model = sonar-pro|https://api.perplexity.ai/chat/completions|PERPLEXITY_API_KEY|8192|0.7

# Mistral AI
ai_model = mistral-large-latest|https://api.mistral.ai/v1/chat/completions|MISTRAL_API_KEY|8192|0.7

# Local model via Ollama (must be running on port 11434)
ai_model = llama3-local|http://localhost:11434/v1/chat/completions|OPENAI_API_KEY|4096|0.7

Adding a Custom Provider

  1. Add the model line to config/aionic.conf
  2. Export the corresponding API key as an environment variable
  3. Restart the server

The model will automatically appear in /v1/models and be available for chat completions.

🌐 API Endpoints

GET /health — Health Check

Returns server health status.

curl http://localhost:8080/health
{
  "status": "ok",
  "timestamp": 1779698657,
  "server": "AIONIC/1.0",
  "version": "1.0.0"
}

GET /v1/models — List Models

Lists all configured AI models.

curl http://localhost:8080/v1/models
{
  "models": ["llama-3.3-70b-versatile", "gpt-4o", "claude-3-5-sonnet-20241022", ...],
  "count": 27
}

POST /v1/chat — Chat Completion

Sends a prompt to an AI model and returns the response.

curl -X POST http://localhost:8080/v1/chat \
  -H "Content-Type: application/json" \
  -d '{"prompt": "What is the capital of France?", "model": "llama-3.3-70b-versatile"}'

Parameters:

  • prompt (required) — The text prompt to send
  • model (optional) — Model name (uses default if omitted)
{
  "response": "The capital of France is Paris.",
  "model": "llama-3.3-70b-versatile",
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 8
  }
}

POST /v1/chat/stream — Streaming Chat

Same as /v1/chat but sends the response as Server-Sent Events (SSE).

curl -N -X POST http://localhost:8080/v1/chat/stream \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Tell me a story", "model": "llama-3.3-70b-versatile"}'

Output format:

id: <request_id>
event: message
data: <chunk>

data: [DONE]

GET /stats — Server Statistics

Returns live server metrics.

curl http://localhost:8080/stats
{
  "requests": 42,
  "responses": 42,
  "active_connections": 1,
  "total_errors": 0,
  "total_ai_calls": 10,
  "uptime_seconds": 3600,
  "bytes_sent": 1048576,
  "bytes_received": 65536,
  "timestamp": 1779698657
}

GET /metrics — Prometheus Metrics

Returns metrics in Prometheus text format for scraping.

curl http://localhost:8080/metrics
# HELP aionic_requests_total Total HTTP requests
# TYPE aionic_requests_total counter
aionic_requests_total 42
# HELP aionic_ai_calls_total Total AI model calls
# TYPE aionic_ai_calls_total counter
aionic_ai_calls_total 10

POST /v1/embeddings — Text Embeddings

Generate vector embeddings for text input.

curl -X POST http://localhost:8080/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"model": "text-embedding-3-small", "input": "Hello world"}'
{
  "data": [{"embedding": [0.025, -0.009, ...], "index": 0}],
  "model": "text-embedding-3-small",
  "usage": {"prompt_tokens": 3}
}

GET /v1/providers — Provider List

Returns the list of configured AI providers.

curl http://localhost:8080/v1/providers

GET / — Root Page

Returns the server welcome page or banner.

curl http://localhost:8080/

🔌 Plugin System

NeuroHTTP supports dynamic loading of shared library plugins at runtime.

Plugin Structure

Each plugin is a .so file in the plugins/ directory with these entry points:

Function Hook Point When Called
plugin_init() — On server startup
plugin_hook(0, ctx) PLUGIN_HOOK_PRE_REQUEST Before routing a request
plugin_hook(1, ctx) PLUGIN_HOOK_POST_REQUEST After routing, before response
plugin_hook(2, ctx) PLUGIN_HOOK_AI_PROMPT Before an AI API call
plugin_hook(3, ctx) PLUGIN_HOOK_AI_RESPONSE After an AI API response
plugin_hook(4, ctx) PLUGIN_HOOK_ON_CONNECT New client connection
plugin_hook(5, ctx) PLUGIN_HOOK_ON_DISCONNECT Client disconnects
plugin_cleanup() — On server shutdown

Building Plugins

# Build all plugins
make plugins

# The .so files are output to build/plugins/

Built-in Plugins

Plugin File Purpose
logstats plugins/logstats.c Logs all requests and AI prompts
openai_proxy plugins/openai_proxy.c Intercepts AI prompts to add system instructions

Creating a Custom Plugin

#define _POSIX_C_SOURCE 200809L
#include <stdio.h>

int plugin_init(void) {
    printf("[myplugin] Initialized\n");
    return 0;
}

int plugin_hook(int hook_point, void *ctx) {
    if (hook_point == 0) {  /* PRE_REQUEST */
        printf("[myplugin] Request received\n");
    }
    return 0;
}

void plugin_cleanup(void) {
    printf("[myplugin] Cleaned up\n");
}

Compile: gcc -fPIC -shared myplugin.c -o build/plugins/myplugin.so

🛠 Makefile Reference

Command Description
make Build production binary (PIE, ASLR, RELRO)
make debug Build with AddressSanitizer + UBSan
make rebuild Clean + full rebuild
make run Build and run the server
make run-debug Build and run debug server
make plugins Build plugin .so files
make test Build and run main test suite
make test-all Run all test suites (unit, fuzz, integration)
make test-moderation Run content moderation tests
make test-gateway Run AI Gateway tests (function calling, embeddings, vision, cost)
make test-cache Run cache tests (in-memory, semantic)
make test-firewall Run WAF tests (IP blocklist, whitelist, rate limits)
make test-fuzz Run fuzz tests (random payloads, edge cases)
make test-prompt-router Run prompt router tests (routing, cost, model management)
make coverage Run all tests with gcov coverage analysis
make check-deps Check system dependencies (hiredis, libcurl, openssl, nghttp2, nasm)
make install Install to /usr/local/bin/aionic
make uninstall Remove installed files
make clean Remove build artifacts
make benchmark Show benchmark command
make docs Generate Doxygen docs
make analyze Run cppcheck static analysis
make format Format code with clang-format
make memcheck Run valgrind memory check
make profile Profile with gprof

🏗 Project Architecture

NeuroHTTP/
├── src/                    # C source files
│   ├── main.c              # Entry point, signal handling, init loop
│   ├── server.c            # HTTP server, connection management, event loop integration
│   ├── iouring_engine.c    # io_uring async I/O engine with epoll fallback
│   ├── tls.c               # TLS 1.3 wrapper (OpenSSL 3.6 / BoringSSL), OCSP stapling, ALPN
│   ├── http2.c             # HTTP/2 session management via nghttp2 (h2 + h2c)
│   ├── router.c            # Hash-table route dispatcher with middleware
│   ├── http_parser.c       # State-machine HTTP/1.1 parser
│   ├── config.c            # Configuration file parser
│   ├── arena.c             # Arena + Slab allocator + ArenaPool
│   ├── ratelimiter.c       # Token-bucket + sliding-window rate limiter
│   ├── observability.c     # Metrics, tracing, provider stats
│   ├── firewall.c          # WAF with SQLi/XSS detection, IP blacklist
│   ├── parser.c            # Legacy HTTP parser (used by older code)
│   ├── stream.c            # SSE streaming support
│   ├── cache.c             # Response cache
│   ├── plugin.c            # Dynamic plugin loader
│   ├── optimizer.c         # Runtime optimizer
│   ├── utils.c             # Utility functions
│   ├── ai/
│   │   ├── prompt_router.c # Multi-provider AI routing (latency/health/cost-aware)
│   │   ├── ai_gateway.c    # AI Gateway: function calling, embeddings, vision, cost tracking
│   │   ├── content_moderation.c # Built-in prompt/response content moderation
│   │   ├── stats.c         # Per-model statistics
│   │   └── tokenizer.c     # Token counting
│   └── asm/
│       ├── crc32.s         # CRC32 (SSE4.2) — hashing
│       ├── json_fast.s     # JSON tokenizer (SSE2/AVX2)
│       └── memcpy_asm.s    # Memory copy (SSE2–AVX-512)
├── include/                # Header files
│   ├── tls.h               # TLS 1.3 + OCSP + ALPN API
│   ├── http2.h             # HTTP/2 session API
│   └── ...
├── config/
│   └── aionic.conf         # Main configuration file
├── plugins/                # Plugin source files
└── Makefile

Data Flow

Client → TCP Accept (plain/TLS)
         │
         ├── TLS 1.3 handshake (if TLS port)
         │   ├── OCSP stapling, ALPN negotiation
         │   └── HSTS headers
         │
         ↓
   Event Loop (epoll / io_uring)
         │
         ├── HTTP/2? ──► h2c upgrade or ALPN h2
         │                   └── nghttp2 session → multiplexed streams
         │
         ↓
    HTTP Parser (state-machine HTTP/1.1 or HTTP/2 frames)
         ↓
      Firewall (WAF)
         ↓
    Rate Limiter (token bucket)
         ↓
      Route Dispatcher
         ├── /health → HealthHandler
         ├── /v1/models → ModelsHandler
         ├── /v1/chat → AI Provider Router
         │                ├── Groq
         │                ├── OpenAI
         │                ├── Anthropic
         │                ├── Gemini
         │                └── ...
         ├── /metrics → PrometheusHandler
         ├── /stats → StatsHandler
         └── /v1/providers → ProvidersHandler
         ↓
   Response → send() / sendfile() / nghttp2 submit_response()
         ↓
   Graceful Shutdown (SIGTERM/SIGINT)
      ├── Stop accepting new connections
      ├── Drain active connections (configurable timeout)
      └── Cleanup resources → exit

📈 Benchmarks

Server Connections Requests/sec Avg Latency Transfer/sec
NGINX 1.29.3 10k 8,148 114ms 1.2 MB/s
NeuroHTTP 10k 2,593 57ms 7.9 MB/s

NeuroHTTP handles heavier AI-rich payloads with lower latency.

🔧 Troubleshooting

Server won't start — "Address already in use"

# Find the process using port 8080
fuser 8080/tcp

# Kill it
fuser -k 8080/tcp

# Or use a different port
# Edit config/aionic.conf: port = 8081

No models available

# Check that API keys are set
echo $GROQ_API_KEY
echo $OPENAI_API_KEY

# Check config file syntax
cat config/aionic.conf | grep ai_model

# Verify the key environment variable name matches
# If the config says OPENAI_API_KEY, then:
export OPENAI_API_KEY="sk-..."

AI request fails

# Check if the provider endpoint is reachable
curl -v https://api.groq.com/openai/v1/models

# Check server logs
tail -f logs/aionic.log

# Verify SSL (try with verify_ssl = 0 in config if you have certificate issues)

Performance tuning

# In config/aionic.conf:
worker_threads = <number_of_cores>   # Match your CPU core count
max_connections = 10000              # Increase for higher load
keepalive_timeout = 60               # Keep connections alive longer
enable_zero_copy = 1                 # Enable for static file serving
enable_smart_routing = 1             # Distribute load across providers

Build fails

# Ensure dependencies are installed
dpkg -l | grep -E "build-essential|nasm|libcurl"

# Clean and retry
make clean && make

# Try without optimization
make debug

📜 License

MIT License — see LICENSE.

🧬 Author

GUIAR OQBA 🇩🇿

"Building the next generation of AI-native infrastructure — from El Kantara, Algeria."

About

High-performance, AI-native server built from scratch in C + hand-written Assembly (SSE2/AVX2/AVX-512). Supports 10 AI providers — OpenAI, Groq, Anthropic Claude, Google Gemini, DeepSeek, Moonshot/Kimi, Zhipu GLM, Perplexity, Mistral AI, and local models via Ollama — all in one binary.

okba14.github.io

Topics

open-source web-server high-performance assembly grpc low-latency neural http3 c-language grpc-server ai-inference ai-native ai-inference-server

Resources

Readme

License

View license

Code of conduct

Code of conduct

Contributing

Contributing

Security policy

Security policy
Activity

Stars

53 stars

Watchers

4 watching

Forks

3 forks
Report repository

Releases

1 tags

Packages

 
 
 

Contributors

Languages