himeshaa

<div align="center"> <img src="app/public/icons/himeshaa-192.png" alt="Himeshaa Logo" width="120" /> # Himeshaa **The Open Source Platform to Build Your Agentic Enterprise** [](https://opensource.org/licenses/MIT) [](https://vuejs.org/) [](https://laravel.com) [](https://www.docker.com/) *Create AI agents. Organize them into teams. Run your agentic enterprise — open source, self-hosted, unlimited.* </div> --- ## 🌌 What is Himeshaa? Himeshaa is the open-source infrastructure for building an **agentic enterprise** — an organization where AI agents and humans work side by side as colleagues across unlimited teams, departments, and workflows. Each agent has a role, expertise, persistent memory, and access to external tools via MCP. They collaborate with your human team in real-time chat and voice, execute multi-step workflows autonomously, hand off work to each other, and escalate when they need human judgment. The **Workspace** is the organization. The **Agent Foundry** is your hiring pipeline. The **Workflows** are your business processes. The **Team Chat** is the open floor where everyone — AI and human — gets work done. Self-hosted. Model-agnostic. 100% open source. Bring your own keys, own every byte. ## 💡 Why Himeshaa? We believe AI agents should work like team members — not as disposable chat windows you close and forget. They should remember context, execute real tasks, hand off work to each other, and collaborate with humans in the same room. We also believe this should be **open source** — built in the open, improved by the community, and owned by whoever runs it. That's what we're building. **The stack:** | Layer | Technology | |-------|-----------| | API | Laravel 12 & Octane (RoadRunner) — persistent workers, no cold starts | | Queues | Redis (7 databases) — 9 named job queues | | Real-time | Laravel Reverb — native WebSocket for agent events, chat, voice | | Frontend | Vue.js 3 + Vue Flow — drag-and-drop workflow canvas | | Storage | PostgreSQL + PgVector — relational data + semantic vector memory | | Files | MinIO (S3) — self-hosted artifact storage | --- ## 🚀 Key Features ### 🤖 Agent Foundry & Marketplace — Hire Your AI Workforce Create, tune, and deploy specialized AI agents from a visual dashboard — each with its own role, system prompt, model provider, temperature, token limits, and knowledge domains. Publish them to an internal **Marketplace** with reviews, ratings, and usage analytics so any team in your enterprise can discover and reuse them. A built-in **Recommendation Engine** surfaces the right agent for the job through collaborative filtering, content-based matching, and trending analysis. ### 🧠 Persistent Memory & Retrieval — Institutional Knowledge Every agent conversation generates memory entries stored in PostgreSQL with vector embeddings (PgVector). Agents retain full context across sessions and retrieve relevant past interactions at inference time. The **Memory Explorer** provides timeline visualization, weekly activity heatmaps, per-agent distribution, and full-text search with configurable granularity. ### ⛓️ Visual Workflow Orchestration — Design Your Business Processes Build multi-agent workflows on a drag-and-drop canvas. The **Graph Execution Engine** processes workflows as DAGs with checkpointed state — supporting start, pause, resume, and recovery from any node. Each agent node decomposes work into subtasks, executes them through the BrainService, generates artifacts, and passes extracted attributes downstream. ### 🧩 Agentic Tool-Use (MCP) Agents call external tools autonomously via the **Model Context Protocol**. The BrainService detects tool-call requests in the LLM response, resolves them through the MCPClientFactory, executes the tool, and feeds results back into the conversation — supporting up to 5 iterative tool-use rounds per request. Data extraction processors parse and structure tool outputs for downstream nodes. ### 🔌 Pluggable Connectors Seven connector types out of the box: * **RESTful API** — configurable HTTP client with auth (API Key, Bearer, Basic), retry logic, payload templates, and intelligent response attribute extraction (`data.users[0].email`). * **GitHub Proactive** — autonomous DevOps intelligence: push event analysis, PR code compliance scoring, issue sentiment analysis, risk assessment, and auto-generated review comments with enriched repository context. * **Slack / Discord** — bidirectional messaging integration. * **Webhook** — inbound/outbound event processing with signature verification. * **Prompter** — direct LLM prompt injection nodes. ### 🧪 AI-Powered Workflow Planner Describe what you want in natural language. The **Workflow Analysis Service** uses the BrainService to generate a structured execution plan — decomposing objectives into orchestrator nodes, tasks, and subtasks. The planning process exposes **Chain-of-Thought reasoning** to the frontend in real time via WebSocket events (`thinking_to_user`), so you see the AI's logic as it designs the workflow. ### 🗣️ Human-AI Collaboration — The Open Floor Real-time chat for each agent (1:1) and **Team Chat** rooms where multiple agents and humans collaborate in the same thread. Speech synthesis events propagate through WebSockets for voice-enabled responses. Every interaction is logged with token counts, latency, and full conversation context. ### 📊 Production Observability * **AI vs. Human performance** comparison over configurable time ranges * **Token usage** breakdown by provider, agent, and workflow * **Latency tracking** with trend indicators and efficiency metrics * **Per-workflow execution logs** with artifact inspection and error recovery monitoring ### 🔐 Workspace & Secrets Vault — Your Departments Unlimited multi-tenant workspaces with role-based access and invitation flows. A dedicated **Secrets & Variables** vault stores API keys, tokens, and environment variables scoped per workflow — keeping credentials out of agent prompts and connector configurations. ### 🛡️ Self-Healing Orchestration The `ErrorRecoveryOrchestrator` catches failed nodes, analyzes the failure, and retries with exponential backoff through pluggable strategies, handlers, and monitors. Long-running workflows complete even when individual nodes encounter transient failures. ### 📝 Template Marketplace Pre-built agent templates with categories, tags, and community ratings. Templates capture the full agent configuration and can be deployed into any workflow in seconds. ### 🔔 Push Notifications Firebase-powered web push for workflow events, task completions, and agent alerts. Configurable per-workflow notification rules with delivery tracking. --- ## 🏗️ Architecture ``` ┌─────────────────────────────────────────────────────────────────┐ │ Vue.js 3 Frontend (PWA) │ │ ┌───────────┐ ┌──────────────┐ ┌───────────┐ ┌───────────┐ │ │ │ Workflow │ │ Agent Chat │ │ Team Chat │ │ Analytics │ │ │ │ Builder │ │ (WebSocket) │ │ (WS) │ │ Dashboard │ │ │ └─────┬─────┘ └──────┬───────┘ └─────┬─────┘ └─────┬─────┘ │ └────────┼───────────────┼────────────────┼───────────────┼───────┘ │ │ │ │ ┌────▼───────────────▼────────────────▼───────────────▼────┐ │ Laravel 12 + Octane (RoadRunner) │ │ ┌─────────────────────────────────────────────────────┐ │ │ │ BrainService (AI Core) │ │ │ │ ┌─────────┐ ┌─────────┐ ┌────────┐ ┌───────────┐ │ │ │ │ │ Gemini │ │ OpenAI │ │ Claude │ │ Ollama │ │ │ │ │ │ │ │ │ │ │ │ (Phi/Qwen)│ │ │ │ │ └─────────┘ └─────────┘ └────────┘ └───────────┘ │ │ │ │ ┌─────────────────────────────────────────────┐ │ │ │ │ │ MCP Tool Execution Loop (5 iterations max) │ │ │ │ │ └─────────────────────────────────────────────┘ │ │ │ └─────────────────────────────────────────────────────┘ │ │ ┌───────────────────────────────────────────────────────┐│ │ │ Orchestration Engine (DAG Executor) ││ │ │ AgentNodeExecutor → ConnectorNodeExecutor → ... ││ │ │ ErrorRecoveryOrchestrator (self-healing) ││ │ └───────────────────────────────────────────────────────┘│ └────┬──────────┬──────────┬──────────┬──────────┬─────────┘ │ │ │ │ │ ┌────▼───┐ ┌───▼────┐ ┌──▼───┐ ┌───▼────┐ ┌──▼────┐ │Postgres│ │ Redis │ │MinIO │ │Reverb │ │Mailpit│ │PgVector│ │9 Queues│ │ (S3) │ │ (WS) │ │(SMTP) │ └────────┘ └────────┘ └──────┘ └────────┘ └───────┘ ``` --- ## ⚙️ Quick Start Getting started is simple — the entire ecosystem runs in Docker. ### Prerequisites * [Docker](https://docs.docker.com/get-docker/) and Docker Compose installed. * An AI model API key (Gemini, OpenAI, Claude, or a local Ollama instance). ### Installation 1. **Clone the repository:** ```bash git clone https://github.com/himeshaa-labs/himeshaa.git cd himeshaa ``` 2. **Configure your AI provider:** ```bash # Edit api/.env and set your preferred AI provider key # Default: GEMINI_API_KEY (Gemini 3.1 Pro) # Supported: OpenAI (GPT-5.5), Anthropic (Claude 4.6/Opus 4.7), DeepSeek-V4, Ollama (Phi-4, Qwen3) ``` 3. **Spin up the Mesh:** ```bash docker compose up -d --build ``` 4. **Access the Application:** | Service | URL | |----------------|---------------------------| | Frontend | `http://localhost:5176` | | API | `http://localhost:8080` | | MinIO Console | `http://localhost:9001` | | Mailpit | `http://localhost:8025` | 5. **Login with the demo account:** | Field | Value | |----------|------------------------| | Email | `admin@himeshaa.local` | | Password | `password` | > The database is automatically seeded on first initialization with this admin user and a curated set of agent templates. --- ## 🧪 Supported AI Models | Provider | Default Model | Type | |------------|---------------------------------|--------------------| | OpenAI | `gpt-5.5` | Cloud API | | OpenAI | `gpt-5.5-instant` | Cloud API | | Anthropic | `claude-4.6-sonnet` | Cloud API | | Anthropic | `claude-opus-4.7` | Cloud API | | Google | `gemini-3.1-pro` | Cloud API | | DeepSeek | `deepseek-v4` | Cloud API (NVIDIA) | | DeepSeek | `deepseek-v4-flash` | Cloud API (NVIDIA) | | Phi | `phi4:14b` | Local (Ollama) | | Qwen | `qwen3:14b` | Local (Ollama) | Switch between providers at any time by changing `AI_DEFAULT_PROVIDER` in your `.env` file. Each provider's model, temperature, max tokens, and timeout are individually configurable via environment variables. --- ## 🤝 Contributing We welcome contributions from the community! Whether you're fixing a bug, adding a new connector, or improving the documentation, please read our [Contributing Guide](CONTRIBUTING.md) to get started. We heavily encourage the use of our **Tactical Lore** commenting standard to keep the codebase engaging and educational. ## 📄 License Himeshaa is open-sourced software licensed under the [MIT license](LICENSE).