Outclaw Documentation
Outclaw is an open-source AI marketing team built and maintained by Outmark. It consists of 21 specialised agents organised into three sub-functions: Product Marketing, Content & Brand, and Growth.
Agents are orchestrated by AlphaClaw, a CMO-level agent that reads your brief, decides which agents to activate, and routes the work through the correct sequence.
Up and running in under five minutes
Option A: Command Center (recommended)
Settings → API → Paste your Anthropic key. We never store it server-side.
Company overview, brand voice guide, ICP doc, competitive summary. Plain text or markdown.
Pick an agent (or let AlphaClaw decide), fill in the brief, hit Run.
Option B: Clone and self-host
git clone https://github.com/outmarkhq/outclaw.git
cd outclaw
cp .env.example .env
# Add ANTHROPIC_API_KEY to .env
npm install
npm run dev
Requires Node 18+. The app runs on localhost:3000 by default. See Self-hosting for production deployment.
AlphaClaw: the CMO agent
AlphaClaw sits at the top of the Outclaw hierarchy. It reads every brief you submit, interprets the goals and context, and decides which agents to activate and in what sequence.
You can bypass AlphaClaw and address a specialist agent directly — but for multi-step work (e.g. "write a product launch plan and then draft the first two blog posts"), AlphaClaw handles the coordination automatically.
Product Marketing, Content & Brand, or Growth?
One agent or a chain? In what order?
Which knowledge base docs are relevant?
Document, plan, brief, or structured data?
GACCS briefs
GACCS is the brief format used across all Outclaw agents. It stands for Goals, Audience, Creative, Channels, Stakeholders. It is not a rigid form — fill in as much or as little as you know. But the more you fill in, the better the output.
Product Marketing agents
These agents handle the strategy layer: positioning, ICP, competitive analysis, messaging, and launch coordination.
Defines your category, competitive frame, and positioning statement. Outputs a structured positioning doc and a one-liner.
positioning, differentiation, category creationBuilds and validates ideal customer profiles. Produces a scored ICP matrix with firmographic and behavioural signals.
ICP, segmentation, personasAnalyses competitors from your provided data or web research inputs. Outputs a battle card and gap analysis.
competitive analysis, battle cards, positioning gapsTranslates positioning into audience-specific messages. Produces a messaging framework by persona and buying stage.
messaging framework, value props, copy architectureProduces a structured product launch plan: phases, channels, timeline, owners, and KPIs.
product launch, go-to-market, launch planWrites battle cards, objection handlers, one-pagers, and demo talk tracks for sales teams.
sales enablement, battle cards, objection handlingAnalyses win/loss data to surface positioning gaps, common objections, and deal pattern trends.
win/loss, deal analysis, pipeline insightsContent & Brand agents
These agents produce the content layer: long-form, social, SEO, email, and visual direction — all grounded in your brand voice.
Codifies your brand voice from existing assets. Produces a voice guide and evaluates new content against it.
brand voice, tone of voice, style guideWrites long-form editorial: blog posts, guides, whitepapers, and reports — in your voice.
blog, long-form, editorial, thought leadershipCreates LinkedIn, Twitter/X, and channel-specific social content. Understands format conventions per platform.
LinkedIn, social media, short-form, engagementStructures content for search: topic clusters, keyword strategy, internal linking, and on-page recommendations.
SEO, keyword research, content clusters, search intentWrites creative briefs for designers and visual assets: mood boards in words, composition notes, and image direction.
visual brief, creative direction, design briefWrites email sequences, nurture tracks, and newsletters. Optimises for open rate, click rate, and reply rate.
email, nurture sequences, drip campaigns, newsletterBuilds editorial calendars, content pillars, and repurposing plans. Connects content to pipeline goals.
content strategy, editorial calendar, repurposingGrowth agents
These agents handle demand generation, pipeline, outreach, and performance — the parts of marketing closest to revenue.
Designs go-to-market motions: channel mix, sequencing, and resource allocation for a given segment or launch.
GTM, channel strategy, market entry, go-to-marketPlans paid campaigns: audience targeting, ad copy angles, budget allocation, and optimisation hypotheses.
paid media, PPC, paid social, ad creative, LinkedIn adsWrites personalised outreach sequences for target accounts. Coordinates with Thiel on account prioritisation.
ABM, outbound, account-based marketing, personalisationReviews pipeline health, stage velocity, and conversion rates. Surfaces where marketing support is needed most.
pipeline, deal velocity, conversion rates, funnel analysisMonitors buying signals: job postings, tech changes, hiring intent, and behavioural cues from target accounts.
intent data, signal monitoring, buying signals, outbound triggersIdentifies and qualifies partnership opportunities. Writes co-marketing briefs and partnership outreach.
partnerships, co-marketing, channel partners, alliancesAudits marketing performance across channels. Produces a prioritised optimisation roadmap.
performance audit, channel ROI, marketing attributionSelf-hosting
Outclaw runs on Node 18+. It can be deployed to any platform that supports Node: Vercel, Railway, Render, or your own server.
Environment variables
ANTHROPIC_API_KEY=sk-ant-... # Required
OUTCLAW_SECRET=... # Used to secure your instance
DATABASE_URL=... # Postgres (optional — defaults to SQLite)
PORT=3000 # Default port
Full deployment guides for Vercel, Railway, and Docker are in the /docs folder on GitHub.
Outclaw defaults to Claude (Anthropic) but the prompt layer is model-agnostic. You can swap in any OpenAI-compatible endpoint by setting:
LLM_PROVIDER=openai
LLM_BASE_URL=https://api.openai.com/v1
LLM_API_KEY=sk-...
LLM_MODEL=gpt-4o
Note: prompt quality is tuned for Claude. Performance on other models may vary. We recommend Claude Sonnet 3.7 or higher for production use.
Contribute to Outclaw
Outclaw is MIT licensed and actively maintained by Outmark. PRs welcome — especially new agent prompts, improved GACCS handling, and integrations.
View on GitHub →