Agentic Marketing Campaign Planner

Intelligent, Autonomous, and Powerful Campaign Strategy Formulation

An AI-native platform designed to orchestrate end-to-end marketing campaigns. Pilot models a realistic campaign-planning workflow: create a project, complete a structured brief, run an autonomous agent with local tool access, approve the final output, and evaluate the specific run against core KPIs.

Stack

• Next.js 16 App Router
• Bun
• TypeScript
• Tailwind CSS
• Zod
• Anthropic Claude Messages API
• OpenCode / OpenAI-compatible fallback
• Mock fallback for offline demos

What It Shows

• Claude-style tool use with local tool execution
• Agent roles: Manager, Research, Strategist, Content, Reviewer
• Lightweight RAG over curated local knowledge
• Prompt-injection and claim-safety guardrails
• Human-in-the-loop approval before final report generation
• Run-specific evaluation for completeness, brief alignment, tool success, guardrails, actionability, and approval readiness
• Browser-local project persistence with localStorage

Product Workflow

1. Campaigns
   • Starts empty on first load.
   • Create or reopen campaign projects saved in this browser.
2. Workspace
   • Complete the structured campaign brief.
   • Run the agent only when the brief has enough usable input.
   • Review tool traces, retrieved knowledge, decisions, blockers, and the approval gate.
   • Approve the draft to create final campaign output.
3. Evaluations
   • Empty until an approved run exists.
   • Scores the selected run against the original brief and generated output.

Quick Start

bun install
bun run dev

Open http://localhost:3000.

Environment

Copy .env.example to .env.local and choose a provider.

AI_PROVIDER=anthropic
ANTHROPIC_API_KEY=...
ANTHROPIC_MODEL=claude-sonnet-4-5-20250929

OpenCode-compatible fallback:

AI_PROVIDER=opencode
OPENCODE_BASE_URL=http://localhost:4096/v1
OPENCODE_API_KEY=...
OPENCODE_MODEL=...

Offline demo:

AI_PROVIDER=mock

If AI_PROVIDER is empty, the app auto-detects in this order:

1. Anthropic when ANTHROPIC_API_KEY exists
2. OpenCode when OPENCODE_BASE_URL and OPENCODE_MODEL exist
3. Mock fallback

API

POST /api/agent/plan

Validates the campaign brief, runs input guardrails, asks the active provider to plan tool calls, executes local tools, and returns a draft.

POST /api/agent/finalize

Requires { draft, approved: true } and returns the final campaign report.

POST /api/evaluations/run

Evaluates a specific run from { brief, draft, report, toolTrace }.

GET /api/evaluations

Returns an empty list in this local-first version. The UI uses run-specific evaluations instead of global dummy scenarios.

Tests

bun test

Covered areas:

• Schema and provider selection
• Local campaign store serialization
• Guardrails
• Tool execution
• Run evaluation scoring

Demo Notes

The project intentionally avoids databases, Redis, vector DBs, and external social APIs. Campaign projects are persisted in localStorage under samy_campaign_projects_v1, so the first visit looks like a real empty product instead of a pre-filled mock dashboard.