AIWeave

TrainWeave orchestrates cost-effective LoRA fine-tuning on AWS EC2 Spot instances through a single Lambda function, eliminating SageMaker overhead and idle compute charges. The tool leverages S3 VPC gateway endpoints to route dataset transfers at zero cost while managing ephemeral training infrastructure that automatically terminates after job completion. By combining spot pricing with intelligent networking, TrainWeave reduces training costs by roughly 50% compared to SageMaker managed training while maintaining full control over the training environment.

TeamWeave is a config-driven multi-agent orchestration platform on AWS that enables teams to define AI pipelines as JSON without requiring code changes to add new workflows or agents. The platform ships with two production-ready patterns—a Visibility Team for content marketing assembly lines and an Improvement Team for personal learning systems—both orchestrated through AWS Step Functions and Bedrock agents. Its distinctive architecture separates control and execution planes, allowing asynchronous pipeline execution with full observability through Amazon Managed Prometheus while keeping all team configurations and artifacts in S3.

TaskWeave is an API-first agent framework built on LangChain and LangGraph that dynamically constructs and executes multi-step workflows from JSON configurations. It supports both explicit task chaining through a `/invoke` endpoint and automatic configuration generation via `/invoke/auto`, where users provide only a question and the system intelligently selects tools from a modular schema registry. The framework excels at orchestrating diverse task types—including LLM prompts, API calls, and data analysis—while maintaining shared memory across tool outputs and gracefully handling missing credentials or network failures with fallback responses.

ToolWeave is a FastMCP server that converts natural-language requests into safe REST API executions by parsing OpenAPI/Swagger specifications and using AWS Bedrock to intelligently plan API calls. It provides a four-tool interface that handles endpoint discovery, parameter extraction, read/write operation execution with proposal gating, and catalog management, all backed by DynamoDB for persistence and mcp-observatory for verification controls. The system ingests API specs from S3, enriches them with metadata, and executes user requests through a secure workflow that immediately runs GET operations while converting write operations into proposals that require explicit commit verification.

ContextWeave is an AWS-native platform that combines GraphRAG and cache-augmented generation to answer deep, evidence-backed questions about developer expertise by ingesting Git repositories, architecture documents, and resume content into a knowledge graph. The system features a semantic response cache that short-circuits the full RAG pipeline for repeated questions, and an adaptive routing layer that automatically classifies documents, recommends optimal chunking strategies, and selects the best retrieval method based on self-improving feedback from Memgraph.

ScreenWeave is an AWS-native platform that crawls websites using Playwright to capture every visual state and interactive transition, storing structured artifacts in S3 for comprehensive analysis. It exposes two independent interfaces—an MCP endpoint for triggering crawls and retrieving screenshots, and a REST endpoint that feeds captured visuals through Claude 3.5 Sonnet via Amazon Bedrock to detect visual anomalies, regressions, and cross-page inconsistencies. The platform supports regression detection across deployments by allowing new QA jobs to reference prior session context, enabling teams to track visual changes over time with AI-powered analysis.

DeployWeave is a modular MCP tool suite that streamlines AWS Bedrock deployments by automating model selection, agent provisioning, and LoRA adapter management with real-time token enforcement. The platform intelligently selects optimal Bedrock models based on latency and cost budgets, provisions up to five agents transactionally with automatic rollback on failure, and manages a dynamic LoRA adapter catalog for fine-tuned model optimization. DeployWeave uniquely separates deployment concerns from orchestration logic, providing an invocation gateway that prevents overspend through atomic token wallet validation and usage deduction on every Bedrock call.

CipherWeave is an agentic cryptography intelligence layer that automatically determines the optimal encryption strategy for AI agents based on data classification, regulatory requirements, and topological risk assessment. The system intercepts key-derivation requests, traverses a graph topology to evaluate risk, and returns a fully justified cipher strategy through a single MCP tool call, with unknown endpoints automatically registered on first use via Bedrock-inferred policy.

MCP Observatory provides a two-phase execution framework for safely handling high-risk MCP tool calls by separating planning from execution through a propose-and-commit pattern. The system evaluates tool proposals for uncertainty and integrity risks before issuing cryptographically signed commit tokens that authorize actual side effects, with comprehensive verification rules including signature validation, expiration checks, and replay protection. It includes generic proposer and verifier wrappers, optional Postgres storage, deterministic fallback responses for blocked operations, and real-world demo scenarios showing end-to-end MCP integration flows.

DataDictionary is an MCP server that manages API field definitions stored in DynamoDB with AI-powered draft generation via Amazon Bedrock. It provides a complete read-write workflow where new data elements are proposed through natural language prompts, verified by MCP Observatory safety gates, and then committed to persistent storage. The tool combines intelligent field generation, proposal verification, and direct query capabilities to streamline the creation and maintenance of API documentation standards.

DeviceWeave is an AI-native execution layer that translates natural language commands into safe, real-time control of IoT devices by combining semantic understanding with enforced policy rules. The system uses a multi-tier resolution approach—starting with fast cosine similarity matching and escalating to Claude Haiku only when confidence falls below threshold—before a dedicated policy engine evaluates every command against stored rules to block unsafe actions or modify parameters before any device I/O occurs.

RoutineWeave is an AI-powered scheduled execution engine that automates prompt-based tasks by running them on cron schedules via Google Gemini. Users define tasks as JSON files with prompts, variables, and schedules, upload them to S3, and the system automatically provisions EventBridge rules to execute them at specified times. Results are delivered directly to email through AWS SNS, with built-in support for dynamic variables, multiple output channels, and observability through CloudWatch.