Skip to main content
MobKit lets you stand up a production multi-agent system in a few lines of code. You describe the agents you want, how they connect, and what operational capabilities they need — MobKit handles the rest: spawning, wiring, routing, lifecycle management, and a live admin console. It builds on Meerkat, which provides the agent runtime (prompt assembly, tool dispatch, sessions, the core agent loop). MobKit adds the orchestration layer that turns individual agents into coordinated groups.
MobKit architecture: Your App → MobKit (orchestration, routing, lifecycle, admin console) → Meerkat (agent loop, sessions, tools, comms). Agent topology: Domain, Personal, and Task agents.

​
What you can build

MobKit’s primary pattern is layered agent architectures — but it’s not limited to any single topology:
  • Domain agents that own a capability (email, calendar, code review, data analysis)
  • Personal agents that represent a user and delegate to domain agents
  • Group agents that coordinate across a team or initiative
  • Task agents that are spawned on demand for short-lived work
You define agents as profiles in a mob config, wire them together with peer connections, and MobKit handles the lifecycle: spawning at startup, retiring when no longer needed, respawning on failure, and reconciling the roster when your desired topology changes.

​
What MobKit provides

  • Orchestration. Bootstrap an entire mob and its operational modules in a single call. Process supervision, health gates, and configurable restart policies.
  • Identity-first continuity. Durable agent identities with roster providers, topology-aware prompts, non-destructive respawn, destructive reset, and in-process reconciliation. Agents survive restarts with stable session IDs.
  • Roster management. List, inspect, spawn, retire, and respawn members. Ensure agents exist before sending them work. Reconcile the roster against a desired state.
  • Operational subsystems. Message routing with sink adapters, risk-based approval gates, cron scheduling with timezone support, and semantic memory — all wired in as supervised modules.
  • Admin console. React web UI with agent sidebar, activity feed, chat inspector, topology view, and health overview — all updating in real time via SSE.
  • Session persistence. JSON file storage for development, BigQuery adapter for production. Complements Meerkat’s own runtime-backed session durability.
  • Authentication. JWT/OIDC validation with JWKS discovery and email allowlists. Google, GitHub, Generic OIDC, and Service Identity providers.
  • SDKs. Python and TypeScript SDKs with typed returns and contract parity tests against the Rust core.

​
MobKit and Meerkat

Meerkat is the engine that makes each agent work. MobKit is the toolkit that makes them work together.

​
Explore the documentation

Quickstart

Bootstrap a runtime with modules and a Meerkat mob in under a minute.

Core Concepts

Modules, events, gating, delivery, scheduling, memory, and sessions.

Guides

Step-by-step walkthroughs for the unified runtime, module system, and console.

Architecture

Internal architecture, crate structure, and design decisions.

APIs

JSON-RPC, REST, and SSE protocol references.

SDKs

Rust, TypeScript, and Python SDK documentation.