Overview
TPS, or Team Provisioning System, is an infrastructure stack for building and operating teams of AI agents. It combines persistent agent identity and memory with team messaging, runtime components, and support for offices distributed across multiple hosts.
The stack is organized into interoperable projects rather than a single monolithic agent. These components cover shared memory, team administration, custom runtimes, and a configurable agent shell.
Key Components
- FLAIR provides signed agent identity, semantic memory, persistent personality, procedures, and relationship or workspace context.
- TPS CLI manages hiring, briefing, dispatch, skills, intra-office mail, and branch offices from one command surface.
- AGENT supplies a mail-driven runtime loop with task turn limits, provider adapters, and process isolation for custom runtimes.
- BOB offers a moldable office-agent shell with identity, memory, mail, channels, scheduling, and role-specific configuration.
Memory and Interoperability
FLAIR runs as a standalone shared-memory layer or as part of a TPS office. It uses in-process embeddings for semantic search and signs reads and writes with each agent’s Ed25519 identity.
The same memory can follow an agent across supported harnesses, including coding CLIs, workflow systems, and agent frameworks. Bridges support memory import and export, while dedicated integrations expose the memory layer through MCP, OpenClaw, and Google Agent Development Kit environments.
Team Architecture
TPS models deployments as a headquarters office connected to agents and optional branch offices. Mail moves work between agents, while encrypted hub-and-spoke tunnels connect offices across hosts. Branch agents can access headquarters memory remotely or operate their own memory service.
The runtime supports multiple model providers and isolates agent processes. BOB can run as a self-restarting user service on macOS or Linux, with in-process scheduling for proactive tasks and pluggable capabilities for communication and observability.
Use Cases
TPS is suited to persistent agent teams whose members need stable identities, shared context, asynchronous work handoffs, and defined roles such as writer, reviewer, coder, QA, or executive assistant. Developers can use the packaged office stack or install the runtime library directly when creating a custom execution environment.