A team of agents, each with its craft.
Each agent masters one field: it reads your data, spots opportunities and proposes actions. The policy engine sorts them, and you approve whatever commits budget.
Meet the team
Each agent has a written scope, a closed list of tools and a cost ceiling.
Polar
Orchestrator
- Has access to
- Your goals, opportunities across every channel, the history of your decisions
- Produces
- It orchestrates every agent: it sets the strategy, arbitrates how the budget is split under your caps and produces the growth plan.
Sirius
Paid Search agent
- Has access to
- Queries, bids, budgets, ads
- Produces
- Exclusions, adjustments, budget proposals
Vega
SEO agent
- Has access to
- Search Console, a crawl of your site
- Produces
- Content and internal-linking opportunities
Altair
AI visibility agent
- Has access to
- Answers from ChatGPT, Perplexity and Mistral
- Produces
- Citation gaps and recommendations
Arcturus
Performance agent
- Has access to
- Spend, conversions, anomalies
- Produces
- Alerts and outcome evaluation
Lyra
Creative agent
- Has access to
- Your brand guide, your best ads
- Produces
- Visuals and variants to review
One agent to orchestrate all the others.
The orchestrator agent produces your strategy and your growth plan. Each task of that plan is sent to the agent concerned, which works within its own scope towards one shared goal: your growth.
Your sources
- Search Console
- GA4
- Google Ads
- AI answers
Polar
produces the strategy and the growth plan
Turns the plan into tasks, handed to the agents concerned, which propose a series of actions
- SiriusTask receivedAnalysingActions proposed
- VegaTask receivedAnalysingActions proposed
- AltairTask receivedAnalysingActions proposed
- ArcturusTask receivedAnalysingActions proposed
- LyraTask receivedAnalysingActions proposed
Proposed actions
→ Policy engine: classifies actions by risk and business impact
- Applied automatically
- Waits for your approval
- Rejected, with a reason
The right rule for every decision.
Deterministic, model call or agentic loop: each agent is set to the best mode for the tasks it runs. Predictable costs and explainable decisions.
Deterministic rule
65 to 70 %
Rules and statistics, zero LLM: thresholds, aggregations, anomalies.
Model call
25 to 30 %
One structured call, validated by a schema, to classify or draft.
Bounded agentic loop
< 10 %
A local loop capped in iterations, cost and duration.
A night of work, read in the morning.
What the agents prepared while you slept. The action log can be inspected line by line and exported when needed, for instance to answer a GDPR request.
- 02:14agent.seaweekend bids −12 %awaiting your approval
- 02:31agent.geo27 citations in ChatGPTadded to the brief
- 03:05agent.studio3 on-brand visualsto review
- 03:48agent.anomalyMeta spend ×3 in 1 hblocked, you are notified
- 04:20agent.seofix brief for /pricingto approve
- 05:02agent.socialcreative v3 fatiguerotation proposed
- 06:15agent.sea3 off-target queriesexclusion proposed
- 07:00reportingmorning brief assembledsent