A planner agent owns the production graph. It decides which specialists run, in what order, and what gets shipped — visibly, with every decision logged.
The critic agent reviews every output against brand voice, tonal constraints, and the brief itself. Disagreements are surfaced; rejections are auditable; the human keeps veto.
Specialists touch the world through bounded tools — generation, rendering, scheduling, publishing. Each tool has a contract. Each call is traced.
Performance data feeds back to retune the brief — not the model. The system learns what works at the brief layer, where humans can read and override.
We sit with your marketing team, follow the assets that ship, follow the ones that do not. Most of MERIDIAN's shape comes out of this first week.
Six specialists with named scopes — ingest, brief, write, render, critique, publish — plus a planner that owns the production graph. Written down before built.
Cloud-native backend, video generation pipelines, Meta API integrations, attribution feedback. Observability from day one.
A marketing operator owns the loop. They can read every brief, override any output, retighten the critic, and roll back any asset.
No. MERIDIAN replaces the production work — the mechanical labor of turning a brief into 14 channel-shaped assets. The interesting work — strategy, brief writing, judgement on what is worth saying — stays human.
The teams we have shipped MERIDIAN to spend more time on strategy after deployment, not less. The system gives them their attention back.
The critic agent enforces it. We start the engagement by writing the brand-voice rules down, in machine-readable form. The critic rejects anything that drifts. Drift is the most common failure mode; the critic exists specifically to catch it.
If the brand voice cannot be written down, that is a different problem and we will say so in the read.
Performance data is summarized weekly and proposed to the human operator as brief-level adjustments — not silent model retraining. The operator approves or edits the proposal; the brief is updated; the pipeline produces against the new brief. The model never silently changes its mind about your brand.
Meta family (Facebook, Instagram, Reels), X, LinkedIn, and a generic webhook for whatever else your stack uses. Email and CRM integration via API. Adding a new channel takes about a week — the planner picks it up once the specialist exists.
For our reference deployments, ~140 assets per week is the steady-state rhythm — half generated, half operator-edited or rejected. Throughput is not the goal; quality through the critic is.
Send a brief. We read the existing workflow, write back an architecture sketch within 48 hours, and propose an engagement. Typical timeline to first production deployment is 8–12 weeks.
Send a brief. We respond in 48 hours with a written read and an architecture sketch. Two solutions per quarter.