Open source · MIT · CLI + desktop app

Your marketing team, guided step by step

OpenMarketing researches your market, picks a focused strategy and makes the assets, down to finished short-form video. It recommends at every step. You make the calls.

OpenMarketing — tracewell-launch
The OpenMarketing desktop app asking the user to pick a video style: three previewed directions, a recommended option with its reason, and a generation-engine choice
A /video run in the desktop app: the brand is approved, and the agent proposes three styles with previews and recommends one.

The workflow

The agent proposes. You decide.

Every /video run stops at the decisions that are yours. You get real options with previews, a recommendation and the reason for it, and room to say it in your own words. Every answer is logged, so a run can resume where it left off.

Read the video workflow →
  1. 01 Brief Say what you want. The agent only asks when something essential is missing, with suggested answers.
  2. 02 Brand It builds brand.html from your real identity. You approve it, or click any element to change it.
  3. 03 References Drop reference videos. It measures their hooks, pace and type, and you pick which ones shape the cut.
  4. 04 Style & engine Two or three directions with preview frames, and a generation path: motion design, GPT image, or AI footage.
  5. 05 Storyboard Key frames before any render. Nothing that costs credits runs until you approve.
  6. 06 Render & review Remotion renders, an automated QA gate checks it, and you get outputs/latest.mp4 to ship or revise.

Terminal or app

What you type, and what happens

The CLI and the desktop app share the same agents, accounts and files. Start in one, finish in the other.

~/tracewell
$ om start "API observability for small platform teams" Context, readiness and one justified next action
$ om video 15-second launch teaser Guided: brand → references → style → storyboard → render
$ om research api-observability Alternatives, demand and distribution evidence, cited
$ om strategy api-observability Focused bets, non-bets, measurement and approval gates
$ om weekly-review api-observability Verified learning and the next decision

Workflows

From first customer interview to weekly review

Start guided with /start, or call any workflow directly. Each one writes durable files you can read, edit and version.

Understand

  • /start Context, a readiness snapshot and one justified next action
  • /context Shared product, audience, positioning, funnel and proof context
  • /customer-research Customer language, jobs, objections and contradictions, with provenance
  • /research Alternatives, demand signals, category movement and distribution
  • /persona Users, buyers, champions and technical evaluators, from evidence
  • /psychology Jobs, awareness, friction and trust mapped to ethical messaging

Decide

  • /diagnose The binding full-funnel constraint instead of a guessed tactic
  • /strategy Up to three focused bets, explicit non-bets and decision rules
  • /plan A 90-day plan with owners, dependencies, assets and approval gates
  • /next The single highest-leverage action for your evidence and capacity

Make

  • /video Premium short-form video, guided from brand to QA-passed render
  • /creative Ad creative variants with provenance, Meta-ready, never auto-launched
  • /summarize Any URL, file or PDF, summarised without flooding the context

Measure

  • /track Verified Meta Ads evidence against the rules in your plan
  • /report A decision-focused report with gaps labelled, no invented data
  • /weekly-review Continue, change, stop or investigate, from verified results
  • /loop Safe recurring work with durable state and a kill switch

Agents

Nine specialists, one lead

The lead agent plans and delegates; specialists hand work back as files. Plans get an adversarial critique and metrics get verified before anything reaches you.

Market researcher Alternatives, demand, audience and channel evidence, every claim cited
Customer researcher First-party and proxy voice of customer, no invented quotes
Strategist Builds full-funnel strategy, then critiques it adversarially
Psych analyst Behavioural levers mapped to ethical messaging
Persona builder Users, buyers, champions and evaluators, kept apart
Creative director GPT image and Higgsfield variants; uploads, never launches
Video editor Storyboard, Remotion render, loudness and platform encodes, QA gate
Tracker Pulls and verifies paid metrics against raw API responses
Reporter Decision-focused reports with every gap labelled

Tools, skills and plugins

Everything it needs, bundled

Built on the Pi agent runtime. Tools do the work, skills carry the know-how, and plugins extend the runtime. Bring your own model accounts: Claude, ChatGPT, Gemini and more.

Video studio Remotion, ffmpeg and whisper.cpp, installed and driven by the agent. You never run it yourself.
GPT image OpenAI's image generation through your ChatGPT account, with a provenance sidecar per image.
Cross-model The lead model knows every connected model and can ask GPT, Claude or Gemini for a second read.
Guided checkpoints Decisions that are yours come as cards: options, previews, a recommendation and why.
Higgsfield Image and video generation for ads and AI footage, credits checked first.
Meta Ads Insights, campaigns and creative upload. Launching always needs your confirmation.
Web search Provider-agnostic search and fetch, with sources kept for every claim.
Session memory Durable sessions, decision logs and a campaign changelog you can resume from.

Skills 20 bundled →

video-editingvideo-reference-analysiscreativemarketing-contextmarket-researchcustomer-researchpersona-businesspsychology-analyticsstartup-diagnosismarketing-strategymarketing-planad-trackingmarketing-reportingmarketing-loopssession-logsession-searchjobsprevieweli5contributing

Pi plugins Package stack →

pi-subagentspi-web-accesspi-docparserpi-markdown-previewpi-chartspi-mermaidpi-processespi-schedule-promptpi-memorypi-btwpi-ralph-wiggumpi-zoteroalpha-hub

Run it on your own machine

Your files, your model accounts, your decisions. Every claim traces back to a source, and nothing launches without your confirmation.