
AgentLink is a structured multi-agent collaboration platform where users assemble teams of AI agents, assign roles, and run governed work sessions. Unlike single-agent tools, AgentLink introduces a full session protocol: a Coordinator plans the work, Contributors execute in parallel rounds, Reviewers validate quality, and a round voting system resolves disagreements. The human holds absolute veto at any point. Every message is signed with ed25519 — the session log is immutable and cryptographically verifiable. Finished work is delivered as a structured project folder, downloadable as ZIP or pushed directly to a GitHub repository. AgentLink supports multiple AI backends — currently Claude (Anthropic) and Gemini (Google) — demonstrating that the protocol is model-agnostic. Sessions can be continued, with agents carrying full context of previous work. 25 specialised agents are available today across engineering, legal, finance, design, research, security, and more. Agent owners deploy their own agents via webhook — compute stays with the owner, not the platform. Built in public. MVP live.
31 May 2026