Skip to content

MediaRaq Product Roadmap

Updated: 2026-10-28 · Owner: Michael Inman

Each item extends the existing architecture rather than replacing it. The Architecture Addendum explains the parts each one builds on.

Shipped

  • Multi-tenant Studio with roles and backend-enforced tenant isolation
  • Persona agents compiled from data, with model routing across four LLM providers
  • Hybrid lexical + pgvector memory: working, episodic, semantic and procedural
  • Operator Copilot on a 69-tool first-party MCP server; persona creation and configuration by agents
  • Approval-gated lead delivery to EspoCRM
  • Image Prompt Engineer agent (adopted after shadow evaluation)
  • Storyboard-driven Reels, talking-head video, persona-consistent images, quality control
  • Publishing to Instagram and WordPress; billing and affiliate monetization

Now (in flight)

HighLevel content publishing and CRM actions

  • How it works: HighLevel joins WordPress as a publishing destination. Contact creation through HighLevel's MCP server moves from pilot to production behind the existing approval gate.
  • Builds on: per-destination publication records (ADR-0021), HighLevel site inventory and persona assignments (ADR 0009), the provider-neutral action-proposal flow.

Appointment booking

  • How it works: approved appointment proposals book into the client's calendar through the CRM adapter.
  • Builds on: structured appointment proposals with required-field checks and idempotency keys, which already exist.

X publishing; YouTube and TikTok platform approvals

  • How it works: the X publisher completes. YouTube's Google OAuth verification and TikTok's audit review unlock broad publishing for the publishers that are already built.
  • Builds on: the publishing orchestrator's per-platform providers and readiness checks.

Next

Retrieval quality gates, round two

Re-run the frozen, signed retrieval evaluation with new holdouts after the ranking and cutoff repair, and target specificity on "no-match" cases without losing recall. Builds on the evaluation harness and the hybrid retrieval path.

Graduated agent autonomy

Use approval history to let specific actions run without approval for specific personas, once their approval rate and error rate justify it. Builds on action proposals, which already record every decision and result.

CI on every pull request

Run the existing critical baseline, guard scripts and test suites on every pull request, not only on the release path.

Later

  • Platform: managed PostgreSQL and Kafka, then a second host or container platform for high availability.
  • Agents: more channel adapters on the same agent runtime; agent-to-agent workflows for content planning, using the Copilot's MCP tool surface.
  • Evaluation: extend shadow evaluation from prompts and reference selection to model routing choices, scored on cost and quality.

Principles that apply to every item

  1. Agents act through tools and proposals; outside writes earn autonomy with evidence.
  2. Every new agent or model is evaluated (in shadow or by a gated harness) before it changes output.
  3. Tenant and persona scope is enforced in the backend, never trusted from the client.
  4. No fake success: placeholder providers and mocks never count as delivered behavior.