How Zevari Works - Zevari Docs
Your agent does the thinking; Zevari does the safe doing. Connect over MCP or REST; every write is approval-gated.
The model
Zevari is the safe LinkedIn execution layer for AI agents. Your agent (Claude, Codex, ChatGPT, OpenClaw, Hermes Agent, or your own code) connects over MCP or a REST API. The agent does the thinking - search, score, draft, sequence - and Zevari does the safe doing, holding your workspace context, Voice DNA, safety rules, and the approval gate.
MCP or REST
Connect over MCP from an MCP client, or over the REST API from your own code. Same capabilities and the same approval gate on both.
The approval gate
Every write (LinkedIn message, connection request, comment, post, email send, campaign launch) is staged for your approval; you approve, then it executes. Reads are not gated. An agent cannot send unattended.
Safety
No browser cookies. Weekly connection ceilings, working hours, and burst caps are enforced server-side.
Hosted state
Zevari holds campaign schedules and the approval queue on its own infrastructure, so sequences advance between your agent's runs.