Concept overview
How Multipliers works
Multipliers is a multi-tenant control plane that orchestrates AI agents across the systems your business already runs on. At its core sits a coordinator loop that, for every agent run, plans a sequence of tool calls, executes them through tenant-scoped connector adapters, evaluates the result against a published spec-card, and emits an MVE outcome that rolls up to the value-chain dashboard.
The platform is deliberately substrate-agnostic. LLM inference can route to Bedrock, Vertex AI, Azure AI Foundry, or direct lab APIs per-tenant. Tool execution can hit customer-hosted MCP servers on Lambda, Cloud Run, or Azure Functions. The orchestrator itself runs on our SaaS by default — or in your VPC via a Helm chart for regulated workloads.
The same codebase serves every customer. Per-client variation lives entirely in data — connector configuration, agent specs, tenant policies, governance thresholds — never in branches or per-client forks. Onboarding a new system means inserting a row through the admin UI, not deploying a new build.
What the full overview will cover
- The orchestrator loop — plan, execute, evaluate, emit
- The agent registry — built-in agents, third-party A2A registrations, Foundry-published custom agents
- The connector runtime — OAuth coordination, retry semantics, idempotency
- The governance dashboard — audit_events, agent_runs, value-chain rollups
- The 5-Contract publish gate — what every agent must satisfy before going live
- The substrate routing layer — how LLM / tool / sidecar / orchestrator route per tenant
- An end-to-end trace of a single run, from trigger to outcome
Coming soon — request the full overview
The architecture deep-dive is being authored with diagrams and end-to-end traces. Email the FDE team to be added to the early-reader list.
Request full overview