Cadenya

Product Introduction

1. Definition: Cadenya is a hosted, model-agnostic agent runtime and orchestration platform. It is a technical layer that sits between your existing APIs/tools and large language models (LLMs), managing the complete agentic loop—from tool discovery and execution to context management and real-time event streaming.
2. Core Value Proposition: Cadenya exists to enable developers and businesses to build, test, and deploy production-ready AI agents quickly and safely, without requiring a complete infrastructure overhaul. Its primary value is in abstracting away the complexity of agent orchestration, allowing teams to focus on defining objectives and connecting capabilities rather than managing low-level plumbing, context windows, or tool-calling logic.

Main Features

1. Unified Tool Layer: Cadenya provides a single integration point for diverse tooling specifications. It natively connects to MCP (Model Context Protocol) servers, ingests OpenAPI specs, and can wrap existing HTTP endpoints. This creates a unified, normalized tool catalog that agents can dynamically discover and use, regardless of the underlying technology.
2. Progressive Tool Discovery & Context Compaction: To optimize token usage and context window efficiency, Cadenya employs a progressive discovery mechanism. Initially, only tool names and descriptions are included in the agent's context. Detailed schemas are fetched only when the agent specifically requests a tool. Combined with automatic context window compaction, this ensures LLM requests remain small and cost-effective.
3. Real-Time Event Streaming & Webhooks: The platform is built for real-time, event-driven integration. Every significant event in the agentic loop—such as tool_approval_requested, sub_agent_spawned, tool_result, or context_window_compacted—can be pushed to your systems via webhooks or Server-Sent Events (SSE). This enables immediate downstream reactions and seamless embedding of agent logic into live applications.
4. Multi-Agent Coordination & Canary Testing: Cadenya supports hierarchical agent structures where a primary agent can spawn sub-agents with different model configurations specialized for specific tasks. It facilitates robust A/B or canary testing by allowing you to run multiple "variations" of an agent (e.g., using Claude Sonnet vs. GPT-5.2) on the same objectives, with performance feedback collected and compared in a unified dashboard.
5. Approval Gates & Safety Controls: For operations requiring human-in-the-loop oversight, tools can be configured as approval-gated. When such a tool is called, the agent pauses, and a tool_approval_requested event is emitted. The action proceeds only after approval is granted via the webhook response, providing a critical safety and compliance layer for high-stakes operations.

Problems Solved

1. Pain Point: The high technical barrier and infrastructure complexity of moving AI agents from prototype to production. Developers struggle with manually managing context windows, tool schemas, token usage, and the statefulness of agentic loops.
2. Target Audience: Platform/ML Engineers building internal agent frameworks; Product Teams in SaaS companies (like the featured Meridian Freight) needing to add autonomous AI features; Developers who want to integrate agentic capabilities without becoming experts in low-level LLM orchestration.
3. Use Cases: Autonomous Customer Support Agents that can execute actions (e.g., rerouting shipments, updating ETAs) using internal APIs. Supply Chain Exception Handlers that monitor logistics and proactively resolve issues like customs holds or delays. Internal Workflow Automations where an agent coordinates across multiple software systems (CRM, ERP, Dispatch APIs) to complete a complex objective.

Unique Advantages

1. Differentiation: Unlike AI SDKs or frameworks (e.g., LangChain, LlamaIndex) which are libraries to be integrated, Cadenya is a fully hosted runtime. Unlike simple chatbot platforms, it is built for complex, tool-calling, multi-step agentic workflows with deep system integration. It competes by offering a "batteries-included" orchestration layer that handles state, memory, and tooling logic out-of-the-box.
2. Key Innovation: Its model-agnostic, provider-agnostic architecture combined with progressive tool discovery. You are not locked into a single LLM provider (use OpenRouter, OpenAI, Anthropic, etc.), and its method of lazy-loading tool schemas is a patented or patent-pending approach to drastically reducing context window waste, which directly lowers cost and improves reliability for agents with large toolkits.

Frequently Asked Questions (FAQ)

1. How does Cadenya handle LLM costs and token usage? Cadenya provides live token metering for cost visibility and employs progressive tool discovery to keep context windows small. By only sending full tool schemas when needed, it reduces wasted tokens on every LLM call, leading to significant cost savings for agentic workflows with many tools.
2. Can I use my existing APIs with Cadenya without modifying them? Yes. A core feature of Cadenya is connecting to your existing stack. You can integrate MCP servers, OpenAPI specifications, or direct HTTP endpoints without rewriting your backend APIs. The platform acts as an adaptor layer that makes your existing capabilities usable by AI agents.
3. What is required for human approval of agent actions in Cadenya? You configure specific tools as requiring approval. When the agent calls such a tool, Cadenya pauses execution and sends a tool_approval_requested webhook to your designated endpoint. Your system (or a human via a UI) reviews the request and sends an approval or denial back, after which the agent proceeds or re-plans.
4. Does Cadenya provide the AI models, or do I bring my own? Cadenya is inference-agnostic. New accounts include credits for OpenRouter, but you must ultimately provide your own API keys for LLM providers like OpenAI, Anthropic, or any OpenAI-compatible endpoint. Cadenya manages the orchestration, not the model inference itself.
5. How can I monitor the performance and outcomes of my agents in production? Cadenya provides a comprehensive observability dashboard. You can track all objectives, view full execution traces including tool calls and token usage, and collect feedback scores on outcomes. This allows for comparing the performance of different agent variations (A/B testing) and iterating on prompts and tooling based on real-world results.