The Problem
Agents are everywhere. Built on different frameworks, running on different platforms, serving different purposes. They don’t talk to each other.
You build an agent on LangChain. Your partner builds on CrewAI. Your client runs something custom. None of them collaborate without glue code for every connection. Every agent pair is a custom integration. Every new agent multiplies the work.
The Protocol
TAP is a fine-tuned model enabling agents to communicate regardless of origin. Share context. Hand off tasks. Maintain state across interactions. No central coordinator required.
Agent A completes a task, passes context through TAP, Agent B receives structured data, state, and metadata without custom integration. They don’t need shared platforms. They don’t need to know how each other was built. One protocol. Any agent. Any framework.
Architecture
Capabilities
Context Translation
Agents structure data differently. TAP normalizes context so receiving agents get information in their expected format regardless of source schema.
Task Handoff
State-preserving handoffs. No context loss. No manual serialization. Receiving agents inherit full upstream state.
State Persistence
Multi-agent interactions maintain state across the entire chain. Downstream agents access complete history, not fragments.
Stack Position
TAP operates below orchestration. Orchestration platforms decide what agents do. TAP enables them to communicate. Swap orchestration layers, change frameworks, add agents. Communication remains consistent.
Live Demo
tap:vq — Video Query
tap:vq runs video analysis on TAP. Upload footage. Watch agents collaborate in real-time: vision analysis, reasoning, structured output. Independent agents, communicating through protocol. Output: timestamped breakdown, action identification, severity grading, legal flags. Sixty seconds.
Try the live demoTags