Getting Started with Nexlem

Nexlem is a Claude Code-native content factory. One command produces a researched, quality-checked blog article plus all derived social media assets — no external services, no per-token billing, just your Claude Max subscription and a local SQLite database.

This guide walks you through installing Nexlem globally, initializing a project with the configuration wizard, and running your first content campaign.

1. Prerequisites

• Bun 1.x — runtime, package manager, and built-in SQLite driver; install from bun.sh: curl -fsSL https://bun.sh/install | bash
• Claude Code — the agent runtime; install via npm install -g @anthropic-ai/claude-code or follow docs.claude.com
• A Claude Max subscription — text generation is delegated to Claude Code subagents; no Anthropic API key is required, no per-token billing applies
• macOS or Linux (Windows via WSL2)

You do NOT need:

• An Anthropic API key
• Any third-party social publishing, image-generation, or AI-gateway service accounts
• A separate database server (SQLite via bun:sqlite is built in)

2. Install

Nexlem is installed globally via your package manager — bun is recommended. See docs/install.md for per-PM commands and verification steps.

3. Configure

Initialize a Nexlem project in your content project directory. Run this inside Claude Code:

/nexlem init

The wizard collects four configuration sections:

The wizard writes three JSON(C) files:

• config/project.jsonc — project-level defaults
• config/business.jsonc — business identity and editorial voice
• config/sites/<slug>.jsonc — per-site settings

All three are files you can edit by hand. Use /nexlem reconfigure to re-run the wizard for any section.

Verify the configuration cascade:

node_modules/.bin/nexlem-sdk query setup.check

A valid output looks like:

{
  "ok": true,
  "checks": [
    { "name": "nexlem_exists",    "status": "pass" },
    { "name": "config_valid",     "status": "pass" },
    { "name": "hooks_installed",  "status": "pass" },
    { "name": "manifest_present", "status": "pass" }
  ]
}

4. Run Your First Campaign

Run inside Claude Code — The steps in this section require Claude Code because Nexlem delegates text generation to Claude Code subagents. The commands in §§ 2–3 above (install, setup, init, setup.check) run without Claude Code; the pipeline steps below do not.

Start a campaign with a topic and your site slug:

/nexlem run "Your article topic" --site <your-site-slug> --type full

The topic is a positional argument to /nexlem run (not a --topic flag). The --topic flag exists only on /nexlem campaign create.

The --type full pipeline has 14 stages:

• Blog channel (stages 1–10, sequential): research → writer → fact-check → eeat → humanizer → seo → disclaimer → image-planner → quality-gate → atomizer
• Social channel (stages 11–14, parallel after atomizer): carousel, caption, reel-script, youtube-short

Nexlem assigns a campaign id and starts stage 1 immediately. Structured log lines appear in Claude Code's output stream:

{"level":30,"time":1747476000000,"event":"agent_start","campaign_id":"...","agent":"research","version":"1.0.0","step":1,"total":14}

Check progress

/nexlem status

Output:

Campaign: Your article topic (<campaign-id>)
Status: running
Progress: 3/14 (research ✓ → writer ✓ → fact-check running)

Review the quality gate

After stage 9 (quality-gate) completes, the gate output at campaigns/<id>/steps/09-quality-gate.json contains:

{
  "verdict": "APPROVED",
  "overall_score": 9.05,
  "min_score_required": 7
}

A score below your configured quality-gate threshold (the quality_gates_business block in config/business.jsonc) produces "verdict": "REJECTED" and halts the pipeline before the social fan-out.

Campaign complete

Campaign: Your article topic (<campaign-id>)
Status: completed
Progress: 14/14
Quality verdict: APPROVED

5. Resume an Interrupted Campaign

Each stage checkpoints to SQLite atomically on success. If a campaign is interrupted, re-run the same command or use the campaign id directly:

/nexlem campaign run <campaign-id>

Nexlem detects completed stages and skips them. Execution resumes from the first incomplete stage.

6. Multiple Sites and Batch Campaigns

Add additional sites to the same project:

/nexlem site add --slug second-site --url https://second-site.com

Queue and run a batch:

/nexlem campaign create --site <slug> --topic "Topic A" --keyword "keyword a"
/nexlem campaign create --site <slug> --topic "Topic B" --keyword "keyword b"
/nexlem run --batch --site <slug>

7. Editorial Calendar

Plan content ahead of time:

/nexlem calendar plan \
  --site <slug> \
  --topic "Top 5 compact hatchbacks" \
  --date 2026-06-01

/nexlem calendar view --site <slug> --weeks 4

8. Inspect Metrics

After a few campaigns, view agent performance metrics:

/nexlem metrics agents --since 2026-05-01
/nexlem metrics campaigns --site <slug>

9. Dev-Mode Logging

For human-readable logs during local development:

bun run src/main.ts campaign run <id> | pino-pretty

Where to Go Next

Edit this page on GitHub