Agent roster

Twelve specialists, one production line.

Each agent owns a narrow mandate, a model and explicit tools. Runs are logged with tokens, cost and score so every draft is auditable.

Run a pipeline
Agents online
11/12
1 idle · health-checked hourly
Total runs
16,400
Across all specialists
Avg. output score
92
Weighted by agent volume
Avg. latency
2.4s
p50 across the pipeline
ResearchWritingQualityGrowth

Cleaning Agent

Normalise raw input
Research

Strips noise from notes, calls, PDFs and web captures. Normalises structure, removes duplicates and flags unverifiable claims.

Text cleanerClaim flagger
Runs
1,284
Avg score
96
Latency
0.8s
GPT-4o mini

Keyword Agent

Extract topics and trends
Research

Maps the cleaned input to core topics, audience keywords and live trend signals. Produces the semantic search brief.

Trend graphKeyword index
Runs
1,192
Avg score
93
Latency
1.1s
GPT-4o mini

Retrieval Agent

Hybrid RAG over your brain
Research

Retrieves the most relevant pieces of your evidence, past posts, call notes and source library using hybrid vector + lexical search.

Vector storeGraph memorySource ranker
Runs
2,081
Avg score
95
Latency
0.4s
text-embedding-3-large

Strategist

Choose the argument
Writing

Turns the research brief into a strategy document: post type, hook style, narrative arc, CTA, tone and proof points.

Format libraryHook matrix
Runs
864
Avg score
92
Latency
2.2s
Gemini 2.5 Flash

Copywriter

Write the first draft
Writing

Writes in your voice, under 1,500 characters, mobile-first. Enforces your cadence, banned words and opening-line rules.

Voice DNABanned wordsFormat playbooks
Runs
1,730
Avg score
90
Latency
4.8s
Claude Sonnet

Reviewer

Score and gate
Quality

Scores hook power, voice authenticity, value density and engagement potential. Sends weak drafts back with specific edits.

Quality rubricAI-pattern detector
Runs
1,688
Avg score
94
Latency
2.9s
GPT-4o

Rewrite Agent

Apply editorial feedback
Quality

Rewrites only what the reviewer flagged, preserving the author’s argument and voice. Iterates until the threshold is met.

Edit diffVoice lock
Runs
611
Avg score
91
Latency
3.4s
Claude Sonnet

Optimizer

Package for reach
Quality

Produces three hook variants, hashtag sets and a copy-ready final. Keeps the substance intact and removes engagement bait.

Hook variantsHashtagsPreview renderer
Runs
1,522
Avg score
89
Latency
1.7s
Gemini 2.5 Flash

Lead Scorer

Qualify the audience
Growth

Scores every engaged profile against your ICP across role, company, buying signal, timing and engagement quality.

ICP graphSignal feed
Runs
3,410
Avg score
94
Latency
0.9s
GPT-4o mini

Profile Auditor

Turn profile into a funnel
Growth

Audits headline, about, featured and experience sections. Generates tested headline formulas and a conversion checklist.

Headline libraryProfile rubric
Runs
286
Avg score
92
Latency
2.6s
GPT-4o

Humanizer

Remove the AI accent
Quality

Detects 24 common AI writing patterns, banned words and uniform rhythm. Proposes human edits without flattening personality.

24-pattern scannerBanned words
Runs
990
Avg score
90
Latency
1.3s
Claude Haiku

Analyst

Close the learning loop
Growth

Synthesises weekly performance into hook patterns, format signals and next best actions. Feeds the strategy layer.

Post analyticsHook library
Runs
742
Avg score
93
Latency
6.1s
GPT-4o
Run history
Every pipeline execution with cost and score
2 logged