Overview
Dispatch is an AI workflow automation platform. Describe a job in plain English — Dispatch turns it into a multi-tool agent, schedules it, runs it, and reports back with a full audit trail.
It is LLM-agnostic, bring-your-own-key, and ships with 25+ first-class integrations across Gmail, Google Calendar, Notion, Slack, GitHub, Linear, Stripe, Discord, X, and arbitrary HTTP endpoints. Triggers can be manual, cron, or webhook — with retries.
The problem it solves
Most automation platforms force operators into visual programming. Most agent frameworks force engineers into bespoke runtime code. Dispatch gives both audiences a single sentence as the entry point — and a real runtime as the exit.
The two technical problems worth naming:
- Plan integrity. A one-line prompt has to compile into a dependency-ordered plan that a runtime can actually execute, retry, and replay — not a chatty transcript.
- Tool surface. Each integration must be addressable by an agent without fragile string parsing. MCP, structured outputs, and per-tool contracts carry the load.
What makes it different
Dispatch is the workflow runtime AI agents actually need: every step is captured, every input and output is replayable, and every credential is the operator's own.
- LLM-agnostic — OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint.
- BYOK keys — Dispatch never stores prompts and never marks up tokens.
- MCP integrations layered on top of a typed tool registry.
- Live audit log with per-step latency and end-to-end replay.
- Cron, webhook, and manual triggers with exponential backoff retries.
Outcome
Dispatch ships as a SaaS at dispatch-wine.vercel.app with a template library covering inbox digests, lead research, support triage, and content repurposing — the loop is real.