Skip to main content

What is meshagent process?

meshagent process is the CLI entry point for running one multi-channel agent. Use it when one agent should be reachable through more than one entry point, but still keep one shared identity, one shared tool configuration, and one shared rules setup. For example, the same agent can:
  • answer in chat
  • respond to email
  • process queue messages
  • be called by another agent as a tool
One meshagent process command creates one agent. If you need separate agents with different rules, tools, or identities, run multiple meshagent process commands.

Why it is useful

meshagent process gives you one place to define the agent and then decide how work can reach it. That brings a few important benefits:
  • one agent identity across multiple entry points
  • one shared set of rules and tools
  • one deployment shape for chat, mail, queue, and toolkit-style access
  • thread continuity when different channels reuse the same thread path

Supported channels today

Today, meshagent process supports these channel formats:
ChannelWhat it means
chatPeople can talk to the agent from MeshAgent Studio, Powerboards, or another chat-style client
mail:EMAILEmail sent to that mailbox becomes work for the agent
queue:NAMEMessages pushed to that room queue become work for the agent
toolkit:NAMEOther agents or participants can call the agent like a tool and get a reply
Each --channel flag adds another way to reach the same agent.

Threads and continuity

meshagent process does not merge all work into one giant live context. The context boundary is still the thread path:
  • turns for the same thread_id are handled together
  • unrelated threads stay isolated
  • continuity only happens when inputs reuse the same thread path
That is what makes cross-channel continuity possible. For example:
  • an email can create or continue a thread
  • later, a chat client can continue that same thread
  • the agent is still the same agent, and the history is still the same thread history

What happens at runtime

At a high level:
  1. a channel receives input from chat, mail, a queue, or a toolkit call
  2. the channel turns that input into a process turn
  3. the runtime routes that turn by thread_id
  4. one thread-scoped execution handles the turn and writes outputs back into the thread model
This is the core behavior behind process: one agent can serve many channels without sharing unrelated thread state.

What meshagent process handles for you

In addition to routing work from channels, meshagent process handles:
  • per-thread routing and queueing
  • streaming text, files, and other outputs into the thread
  • reasoning summary updates when the selected adapter emits them
  • tool-call lifecycle state such as pending, running, logs, completion, and failure
  • tool approvals and resume flows
  • interrupts and steering for active turns
  • thread status and pending-message metadata for clients
Some model behavior still comes from the configured LLM adapter rather than from meshagent process itself.

Under the hood

Under the hood, meshagent process assembles a small runtime out of a few core pieces:
  • SingleRoomAgent: the room-connected agent shell that holds the shared identity, requirements, rules, and tool configuration
  • channel adapters such as ChatChannel, MailChannel, QueueChannel, and ToolkitChannel: the pieces that turn chat, email, queues, and toolkit calls into process messages
  • AgentSupervisor: the router that receives channel messages and sends each turn to the correct thread
  • LLMAgentProcess: the per-thread execution loop that runs one active thread at a time
  • AgentProcessThreadAdapter: the bridge that writes outputs, tool state, and turn lifecycle updates back into the thread model
For the full implementation model, see Multi-channel Agent Architecture. For the thread model those runtime pieces write into, see Threads Overview.

Where to go next