Dograh is an open-source voice AI platform for building, deploying, and monitoring conversational voice agents. It provides a visual workflow builder (React/Next.js frontend) where users compose nodes (start, agent, end-call, tools, integrations) into call-handling graphs. The backend (Python/FastAPI) orchestrates real-time voice pipelines using the Pipecat framework, supporting multiple LLM, TTS, and STT providers (OpenAI, Google Gemini, Deepgram, ElevenLabs, etc.) and telephony carriers (Twilio, Vonage, Telnyx, Plivo, Asterisk ARI, WhatsApp, etc.). The system handles inbound/outbound calls, campaign management with traffic splitting and retry logic, text chat sessions, and embeddable widgets. It includes an MCP server that allows AI coding assistants to programmatically create and edit workflows, and auto-generated Python and TypeScript SDKs for programmatic access. The platform is multi-tenant with organization isolation, usage tracking, billing hooks, and extensive observability (Langfuse tracing, PostHog analytics, Prometheus metrics, call event export to BigQuery). Architecturally, the backend follows a modular provider registration pattern: telephony providers, LLM/TTS/STT services, and integration packages (Noveum, Tuner, Paygent) self-register via import side effects, enabling easy extensibility. The workflow engine uses a node-spec system that drives both the UI form renderer and SDK code generation, ensuring consistency. The system uses PostgreSQL (with pgvector for embeddings), Redis (for caching, pub/sub, rate limiting, task queue), and object storage (S3/MinIO). Deployment supports Docker Compose, Helm/Kubernetes, and a devcontainer setup. Key entry points include the FastAPI application (api/app.py), the MCP server (api/mcp_server/server.py), the ARQ background worker (api/tasks/arq.py), and the telephony ARI manager (api/services/telephony/ari_manager.py). The frontend is a Next.js application with a workflow editor, run viewer, settings pages, and billing dashboard. The SDKs provide typed workflow builders and client libraries for both Python and TypeScript.
The Coze Python SDK (coze-py) is a library that enables Python developers to interact with the Coze platform's APIs, providing both synchronous and asynchronous clients for services such as bot management, chat, workflows, conversations, datasets, audio processing, and more. It handles authentication (including PAT, OAuth variants), makes HTTP requests via httpx, and processes responses using Pydantic models for validation. The SDK follows a modular architecture with a main client class (`Coze`/`AsyncCoze`) that provides access to service-specific clients via properties. Key reusable components include a pluggable authentication system (supporting TokenAuth, JWTAuth, OAuth flows), a generic pagination system for list operations, streaming response handlers for Server-Sent Events and WebSocket messages, a centralized HTTP requester abstraction, and base classes for WebSocket clients. These components demonstrate non-obvious craftsmanship and could be extracted for use in other API clients. The SDK is actively maintained, as evidenced by recent commits and the deprecation of outdated modules in favor of newer alternatives (e.g., `cozepy.datasets` replacing `cozepy.knowledge`). However, it exhibits medium technical drift due to tight coupling to the Coze API (lacking API version abstraction) and potential susceptibility to changes in authentication flows or WebSocket dependencies. Despite these risks, the codebase maintains high quality through type hints, comprehensive testing, linting, and pre-commit hooks.
Lumen is an open-source agent framework for chatting with data and building interactive dashboards, built on the HoloViz ecosystem (Panel, hvPlot, Bokeh). It provides a dual interface: a YAML-based declarative dashboard specification system for non-AI use cases, and an AI-powered natural language interface that generates SQL queries and Vega-Lite visualizations from user prompts. The architecture is modular, with clear separation between data sources (files, SQL databases, REST APIs, xarray), data pipelines (filters, transforms, SQL rewrites), views (hvPlot, Vega-Lite, DeckGL, tables), and an AI agent layer (Planner, Coordinator, specialized agents for SQL, visualization, validation, and source ingestion). The AI system supports multiple LLM providers (OpenAI, Anthropic, Google, Ollama, etc.), vector stores for RAG (Numpy, DuckDB, ChromaDB), and a tool system for external API integration. The codebase demonstrates high craftsmanship with reusable patterns like spec-driven component creation (MultiTypeComponent), composable SQL transforms, and a consistent agent/tool abstraction. While primarily an application framework, it includes data pipeline components that could be extracted for data engineering use. The project is actively maintained with extensive test coverage, though its large optional dependency surface and rapid feature velocity introduce some maintenance risk.
This repository contains the Microsoft 365 Agents SDK and samples for building conversational agents deployable to Microsoft 365 channels (Teams, Copilot Studio, web chat) and custom channels. The SDK is AI-service agnostic, allowing integration of any AI model or service. Samples demonstrate common agent patterns including welcome messages, tool usage (dice roller, inventory, weather lookup), streaming responses, authentication (OBO, auto sign-in), observability (OpenTelemetry), and integrations (Genesys, Slack, Semantic Kernel, LangChain). The architecture follows a consistent pattern across languages: initialize storage, connection manager, adapter, and authorization; instantiate AgentApplication; register event handlers; start server. Entry points vary by language (Program.cs for .NET, src/index.ts for Node.js, src/main.py for Python). External dependencies include Microsoft Graph, GitHub API, Azure OpenAI, and various agent frameworks.
This repository defines the Conversational Finance Governance Framework, an open specification (v0.4) for embedding runtime authority controls into conversational AI agents operating in fintech contexts. It includes a reference implementation (Lite sidecar) that demonstrates the control plane: classifying utterances along Financial Commitment Risk (F0-F3) and Regulatory Sensitivity (R0-R3) axes, binding each turn to an Action Object (inform/advise/commit/transact/represent), applying a decision table (FREE/AUTH/ESCALATE), enforcing classification assurance, issuing scoped credential records, and producing a hash-chained audit trail. The sidecar is a CLI tool that takes a product-line mapping YAML (payments, lending, insurance, collections) and an utterance, then outputs the governance decision as JSON. It does not call an LLM or execute real transactions; it is a control-plane demo. The repository also contains the full specification documents (v0.3 and v0.4), reference mapping files, and a test suite with gold-standard classification fixtures. The architecture is a monolithic CLI with a clean separation between classification, decision logic, and audit logging. The code is well-structured, uses dataclasses, and includes a reusable hash-chained audit log component. The last commit date (2026-09-09) appears to be in the future relative to the current date, which is anomalous but taken at face value indicates active maintenance.
Eidolon is an open-source framework for building, deploying, and managing multi-agent AI applications. It provides a Python SDK with abstractions for LLM integration (OpenAI, Anthropic, Mistral, Ollama, Azure), retrieval-augmented generation (RAG) pipelines, memory systems (vector stores, symbolic memory, file storage), and document processing (loaders, parsers, splitters). Agents are configured declaratively via YAML and can communicate with each other, use tools, and stream responses. The framework includes a Kubernetes operator for production deployment, a web UI built with Next.js and React for chatbot interaction and developer tooling, and a TypeScript client library. It also features built-in agent types like SimpleAgent, RetrieverAgent, SQLAgent, and a Tree-of-Thought agent, along with logic units for web search, browser automation, and API integration.
IBM TJBot is a Raspberry Pi-based robotics application that interacts with IBM Watson AI services through various 'recipes'. Each recipe demonstrates a different AI capability: conversation (chatbot using Watson Assistant, Speech to Text, and Text to Speech), sentiment analysis (monitoring Twitter tone and changing LED color), speech to text (voice-controlled LED), and translation (speech-to-speech language translation). The architecture consists of a core TJBot library that abstracts hardware and service interactions, with each recipe being a standalone Node.js application. Key entry points are the recipe-specific JavaScript files (conversation.js, sentiment.js, stt.js, translator.js) and the bootstrap script for Raspberry Pi setup. Data flows involve capturing input (audio, Twitter stream), processing through Watson services, and triggering hardware actions (LED, servo, speaker) based on results.