Skip to content

Releases

Release v0.5.0

Release v0.5.0 Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 21 Aug 17:15
v0.5.0
a3a2c6c

Full Changelog: v0.4.1...v0.5.0

Full Changelog: v0.4.1...v0.5.0

Assets 2
Loading

Nexus v0.4.1 - Universal AI Coding-Agent Gateway

Nexus v0.4.1 - Universal AI Coding-Agent Gateway

Choose a tag to compare

View all tags
@rachidSabah rachidSabah released this 14 Aug 15:13
v0.4.1
df6b4f0

What is Nexus?

Nexus is a universal AI coding-agent gateway and autonomous control plane that dynamically discovers models across providers and exposes them through one OpenAI-compatible and Anthropic-compatible local endpoint (http://127.0.0.1:8787).

Architecture

nCODING AGENTS -> NEXUS GATEWAY -> UNIVERSAL MODEL FABRIC -> AUTONOMOUS ROUTING -> PROVIDERS -> MODELS USER -> NEXUS -> APPLICATION ENGINE -> WORKFLOW -> AGY -> NEXUS MODEL FABRIC -> PROVIDER -> MODEL n

Highlights & Verified Capabilities

  • Dynamic Model Discovery: 571 models discovered dynamically across Nvidia NIM, Mistral, OpenRouter, and Cerebras.
  • Coding-Agent Proxy: Live inference verified for Claude Code, Codex CLI, Hermes CLI, and AGY with zero credential leakage.
  • AGY Application Builder: Autonomous plan, risk analysis, workspace sandboxing, implementation, and verification tested to completion.
  • Resilient Key Vault: AES-256-GCM encryption with per-request rotation, 401 isolation, and 429 cooldown.
  • Token Optimization: Prompt compression and tool-schema normalization active with measured savings.
  • Mission Control Dashboard: Next.js 15 UI with zero-flicker dark/light themes and single vertical scroll layout.

Installation

Windows (PowerShell):
powershell irm https://raw.githubusercontent.com/rachidSabah/codingghosts/main/scripts/install.ps1 | iex

Linux / WSL / macOS:
bash curl -fsSL https://raw.githubusercontent.com/rachidSabah/codingghosts/main/scripts/install.sh | bash

Loading

Agent Nexus OS v0.4.0 — Agent Orchestration Platform

Agent Nexus OS v0.4.0 — Agent Orchestration Platform

Choose a tag to compare

View all tags
@rachidSabah rachidSabah released this 05 Aug 12:10
v0.4.0
2b722ef

Agent Nexus OS v0.4.0 — Agent Orchestration Platform

Phase 4 of Agent Nexus OS. Evolves the gateway into a complete self-hosted AI agent orchestration platform while preserving the existing hexagonal architecture, DDD, event-driven design, and plugin-first principles.

What's New

7 new packages

Package Purpose
@anx/agents Agent Registry — register, discover, monitor 10 built-in agents
@anx/runtime Agent Runtime — lifecycle, sessions, retry, timeout, streaming
@anx/workflow Workflow Engine — define, version, execute, pause/resume/replay
@anx/memory Memory system — short-term + long-term with vector stores
@anx/tools Tool Management — permissions, sandboxing, discovery, logging
@anx/task-router Intelligent AI Task Router — 5-stage planning pipeline
@anx/persistence Repository pattern — PostgreSQL, SQLite, Redis adapters

Extended packages

  • @anx/core — 16 new domain event types (agent.*, workflow.*, memory.*, tool.*, team.*)
  • @anx/a2a — TeamManager with teams, proposals, voting (with power weighting), consensus, shared workspaces
  • @anx/observability — 15 new metrics: agent_execution_time_ms, workflow_duration_ms, anx_agent_tokens_used_total, anx_workflow_cost_usd_total, success rate

Built-in agents (10 templates)

claude-code, codex-cli, gemini-cli, hermes-cli, opencode, openhands, aider, continue, mistral-coder, deepseek-coder

Built-in workflows (3 templates)

  • software-development-pipeline — architecture → implement → review → test → document
  • bug-triage — reproduce → diagnose → fix → verify
  • code-review — security + performance + style (parallel) → consensus

Built-in tools (12 definitions)

filesystem.read/write/list, terminal.execute, git.status/commit/diff, browser.navigate/screenshot, database.query, http.fetch, mcp.invoke

Intelligent AI Task Router

5-stage planning pipeline:

  1. TaskClassifier — classifies requests into 12 task types
  2. CapabilityMatcher — maps task types to required capabilities
  3. AgentSelector — scores eligible agents (tags, status, concurrency, cost)
  4. ModelSelector — picks the best model per task type per agent
  5. ExecutionPlanner — assembles single-step or multi-stage plans

Example: "Build a SaaS application" → 5-step plan:

  • architecture → claude-code / claude-3-5-sonnet
  • backend → deepseek-coder / deepseek-coder
  • frontend → gemini-cli / gemini-1.5-pro
  • testing → codex-cli / gpt-4o
  • documentation → mistral-coder / mistral-large

Multi-Agent Collaboration

  • Teams — form teams with named roles (planner, executor, critic, observer)
  • Proposals — create proposals, vote (yes/no/abstain)
  • Voting power — weighted votes per member
  • Consensus — auto-close proposals when all members vote
  • Shared workspace — artifacts (document, code, decision, review, note)
  • Task delegation — route tasks to specific team members via A2A

Memory System

  • Short-term: in-memory, per-session, TTL-based expiry
  • Long-term: vector-indexed, survives restarts
  • Adapters: InMemoryVectorStore (default), Qdrant, Chroma, pgvector (stubs for v0.5)
  • Embeddings: GatewayEmbeddingsProvider (uses /v1/embeddings), FakeEmbeddingsProvider (tests)
  • Memory interface: store / search / delete / summarize / get / list

Tool Management

  • Permission policies: allow / deny / ask
  • Wildcard matching: filesystem.* matches filesystem.read
  • Per-tool JSON Schema for input validation
  • Execution log with filtering by agent / tool
  • Sandboxing via AbortSignal timeouts

Persistence Layer

  • Repository pattern — business logic never touches DBs directly
  • InMemory (default), SQLite (via node:sqlite), PostgreSQL (stub), Redis (stub)
  • createPersistence() factory picks backend from config

25+ new gateway endpoints

GET    /v1/agents                          List registered agents
GET    /v1/agents/stats                    Aggregate agent stats
GET    /v1/agents/:id                      Get one agent
POST   /v1/agents                          Register a new agent
DELETE /v1/agents/:id                      Unregister an agent
POST   /v1/agents/:id/tasks                Execute a task on an agent
GET    /v1/workflows                       List workflow definitions
POST   /v1/workflows                       Create a new workflow (or new version)
GET    /v1/workflows/:id                   Get a workflow (latest or specific version)
POST   /v1/workflows/:id/execute           Start a workflow execution
GET    /v1/workflows/:id/executions        List executions for a workflow
GET    /v1/workflows/:id/executions/:eid   Get execution status
POST   /v1/workflows/:id/executions/:eid/pause
POST   /v1/workflows/:id/executions/:eid/resume
POST   /v1/workflows/:id/executions/:eid/cancel
POST   /v1/workflows/:id/executions/:eid/replay
POST   /v1/plan                            Generate an execution plan
POST   /v1/memory/:namespace/store         Store a memory
POST   /v1/memory/:namespace/search        Semantic search
GET    /v1/memory/:namespace/list          List memories
DELETE /v1/memory/:id                      Delete a memory
GET    /v1/tools                           List registered tools
POST   /v1/tools/:name/execute             Execute a tool
GET    /v1/tools/log                       Query execution log
GET    /v1/teams                           List teams
POST   /v1/teams                           Form a new team
GET    /v1/proposals                       List all proposals

Dashboard upgrades (5 new pages)

  • /agents — agent registry table with status, capabilities, models, in-flight tasks, cost multiplier
  • /workflows — visual workflow builder showing step flow + execution history with per-step status bars + pause/resume/cancel/replay buttons
  • /teams — team list, proposals with vote counts (yes/no/abstain), shared workspace artifacts
  • /memory — semantic search interface, recent memories, namespace filter, scope pills
  • /marketplace — plugin / agent / tool catalog with install buttons

Tests

  • 100+ new Vitest tests across all 7 new packages + extended a2a
  • Playwright E2E tests for all 5 new dashboard pages + 6 API endpoint tests
  • k6 load test plan targeting 100 concurrent workflows, 50 agents, 500 WebSocket connections

Documentation

  • docs/AGENT_DEV.md — how to build, register, and operate AI agents
  • docs/WORKFLOW.md — how to define, execute, and replay multi-agent workflows
  • docs/README.md — updated to include new guides

Architecture (preserved)

All Phase 4 features integrate through existing ports and adapters:

  • Hexagonal Architecture ✅
  • Domain Driven Design ✅
  • Event-Driven (16 new event types) ✅
  • Plugin-First ✅
  • SOLID ✅
  • 100% TypeScript strict ✅

No existing modules were rewritten or simplified. All Phase 1-3 functionality (provider abstraction, MCP, A2A, plugins, routing, dashboard, security, observability) is preserved.

What's Next (Phase 5)

The logical next milestone is Autonomous Agent Swarm + Self-Improving AI Development Loop:

  • Agent swarms that self-organize without human-defined teams
  • Self-improving workflows that learn from past executions
  • Continuous evaluation and re-ranking of agents per task type
  • Autonomous code review and merge
  • Self-healing infrastructure

Installation

# Docker
docker compose up

# Native (Node 22+)
pnpm install
pnpm build
pnpm --filter @anx/gateway start

Documentation

License

Apache-2.0

Full Changelog: v0.1.0...v0.4.0

Loading

Release v0.1.0 — Agent Nexus Gateway

Release v0.1.0 — Agent Nexus Gateway

Choose a tag to compare

View all tags
@rachidSabah rachidSabah released this 05 Aug 11:26
v0.1.0
18cbb6f

Agent Nexus Gateway v0.1.0

The first public release of Agent Nexus Gateway — the most advanced local AI Gateway. Universal routing, streaming, MCP, A2A, plugins, and a beautiful dashboard.

Highlights

Universal AI Gateway

  • OpenAI-compatible REST API (/v1/chat/completions, /v1/embeddings, /v1/models)
  • SSE streaming
  • WebSocket (real-time dashboard feed)
  • MCP server (expose gateway tools to Claude Code, Continue, Cline, …)
  • MCP client (aggregate external MCP servers)
  • A2A protocol (Agent-to-Agent coordination)
  • Plugin framework with 9 lifecycle hooks

17 Provider Adapters

OpenAI, Anthropic, Google Gemini, OpenRouter, DeepSeek, Mistral, xAI (Grok), Groq, Together, Fireworks, Cerebras, Cloudflare AI, Ollama, vLLM, LM Studio, LiteLLM, Azure OpenAI.

8 Routing Strategies

weighted, round_robin, least_latency, least_cost, highest_quality, capability_match, priority, budget_aware — plus circuit breakers, automatic failover, retries, and health monitoring.

Dashboard

Next.js + TypeScript + Tailwind dashboard with real-time metrics, latency charts, token usage, cost analytics, provider health, request logs, audit log, and network diagnostics.

Security

AES-256-GCM encrypted credential vault, RBAC with wildcard permissions, JWT (HS256), audit logging, Zero Trust architecture.

Observability

OpenTelemetry interface, Prometheus /metrics endpoint, structured JSON logs, event bus bridge.

Networking

HTTP/HTTPS/SOCKS5 proxy support, DNS-over-HTTPS, IPv4/IPv6, latency measurement, connection diagnostics, automatic failover between proxies.

Deployment

Docker (multi-stage, ~150MB image), Docker Compose with optional Prometheus + Grafana, Kubernetes manifests (HPA, PDB, Ingress, headless service), Helm chart.

Developer Experience

CLI (anx), TypeScript SDK, native integrations with Claude Code, Continue, Cursor, Cline, Roo Code, Aider, Zed, JetBrains, Neovim, Emacs, and more.

Architecture

  • 100% TypeScript strict
  • Hexagonal architecture (ports & adapters)
  • Domain-Driven Design
  • SOLID principles
  • Event-driven
  • Plugin-first

Honest Design

Agent Nexus Gateway never relies on undocumented APIs or techniques intended to evade provider usage limits. It maximizes reliability through legitimate means: intelligent routing, retries, failover, caching (planned), and user-configured providers and networking.

Installation

Docker

git clone https://github.com/rachidSabah/codingghosts.git
cd codingghosts
export OPENAI_API_KEY=sk-...
docker compose up

Native (Node 22+)

corepack enable
pnpm install
pnpm build
pnpm --filter @anx/gateway start

Installer scripts

# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/rachidSabah/codingghosts/main/scripts/install.sh | bash

# Windows (PowerShell)
iwr -useb https://raw.githubusercontent.com/rachidSabah/codingghosts/main/scripts/install.ps1 | iex

Documentation

What's Next

See ROADMAP.md. The 0.2 release will add semantic caching, prompt cache passthrough (Anthropic cache_control), OpenTelemetry SDK integration, and Grafana dashboards.

License

Apache-2.0

Full Changelog: https://github.com/rachidSabah/codingghosts/commits/v0.1.0

Loading