Agentset.ai

Accepts 22+ file formats (PDF, DOCX, XLSX, PNG, EML, etc.) and URLs via SDK, automatically parses content into structured text, applies configurable chunking strategies, and attaches custom metadata per document. The ingestion pipeline processes files asynchronously with job status tracking, enabling bulk document onboarding without blocking application flow. Supports multimodal content including images, graphs, and tables with native extraction capabilities.

Converts user queries into vector embeddings and performs similarity search across indexed documents, optionally filtering results by metadata predicates before retrieval. A reranking layer (algorithm unspecified) refines result precision after initial semantic matching. Supports hybrid search combining semantic and traditional retrieval mechanisms, though the hybrid implementation details are undocumented. Returns ranked results with relevance scores and source attribution.

Provides logging and observability features for tracking ingestion progress, search performance, RAG generation quality, and system errors. Logs include request/response traces, latency metrics, token usage, and error details. Observability data is accessible via API and optional dashboard for monitoring system health, identifying bottlenecks, and debugging issues. Supports integration with external monitoring platforms (DataDog, New Relic, etc.).

Offers three pricing tiers with different feature sets and usage limits: Free tier (1,000 pages, 10,000 retrievals/month, no connectors), Pro tier ($49/month, 10,000 pages included, unlimited retrievals, per-connector charges), and Enterprise tier (custom pricing, BYOC/self-hosted, unlimited pages, custom features). Usage is measured in 'pages' (1,000 characters = 1 page) rather than documents, enabling predictable cost scaling. Connector costs ($100/month each on Pro) are separate from base subscription.

Chains semantic search results directly into an LLM prompt, grounding generated responses in retrieved documents. Automatically tracks and attributes citations to source documents, enabling end-users to inspect the evidence backing each answer. Supports pluggable LLM providers (OpenAI, Anthropic, Google, xAI, Azure, Cohere, Qwen, Mistral, DeepSeek) via configuration, abstracting provider-specific APIs. Reduces hallucinations by constraining generation to indexed knowledge.

Extends simple RAG with AI-driven planning and multi-hop retrieval, enabling the system to decompose complex queries into sub-questions, retrieve relevant documents iteratively, and reason across multiple sources. Integrates with Vercel's AI SDK for agent orchestration, allowing the LLM to decide when to search, what to search for, and how to synthesize results. Supports custom tool definitions and agentic reasoning loops without manual prompt engineering.

Automatically syncs documents from external data sources (Google Drive, SharePoint, Notion) into Agentset namespaces via pre-built connectors. Handles authentication, incremental updates, and metadata extraction from source systems. Connectors are charged per-connector on Pro tier ($100/month each), enabling organizations to maintain live links between source systems and RAG indexes without manual re-ingestion. Webhook events notify downstream systems of sync completion.

Generates shareable preview links to chat interfaces for RAG responses, enabling end-users to interact with grounded answers without accessing the backend system. Interfaces are customizable (branding, instructions, model selection) and collect user feedback (thumbs up/down, comments) for quality monitoring and model improvement. Feedback data is stored and accessible via API for analytics and fine-tuning workflows.

Exposes Agentset RAG capabilities as an MCP server, enabling external applications (Claude, other AI agents, custom tools) to invoke semantic search and RAG operations without direct API calls. MCP standardizes the interface for tool use, allowing Agentset to be plugged into any MCP-compatible client. Supports function-calling semantics with schema-based tool definitions for search, retrieval, and chat operations.

Enables enterprise customers to deploy Agentset infrastructure on their own cloud accounts (AWS, GCP, Azure) or on-premises, maintaining full control over data residency, infrastructure, and compliance. BYOC deployments use customer-managed vector databases (Pinecone, Qdrant) and compute resources, eliminating data transfer to Agentset infrastructure. Self-hosted option provides complete source code and deployment automation for air-gapped or highly regulated environments.

Abstracts LLM provider differences (OpenAI, Anthropic, Google, xAI, Azure, Cohere, Qwen, Mistral, DeepSeek) behind a unified configuration interface, enabling model selection and switching without code changes. Handles provider-specific authentication, API formats, and response parsing transparently. Supports model-specific features (function calling, vision, streaming) while maintaining consistent application-level semantics.

Emits webhook events for key system events (document ingestion completion, sync status, feedback collection) to customer-specified endpoints, enabling event-driven downstream workflows without polling. Webhook payloads include event metadata (timestamp, namespace, status, error details) for routing and error handling. Supports retry logic and delivery guarantees for reliable event propagation.