Skip to content

VIT TEAM Academy

Agents

Every agent is a Claude model plus a playbook: its role, rules and know-how distilled from open-source GitHub repositories with the best marketing, sales and SEO practices. The knowledge is attached to the agent automatically in every answer.

How the team learns

We took the best ideas from open-source agent projects on GitHub and rebuilt them for business. The team doesn't start from zero every day: it remembers, learns from your decisions and researches the internet on its own.

01

A self-check before handing over

Every employee has its own quality checklist: the content maker checks the hook, specifics and call to action, the hunter checks verified contacts and a reason to reach out. Before handing work over, a strict checker compares it with the checklist and your lessons, and the employee fixes what failed. The card shows it: "Self-check 7 of 8, 1 fixed".

Based on Self-Refine, CRITIC, Prometheus

02

A personal notebook for each

Each employee keeps its own records: the hunter a lead base, the content maker its topics, the SEO specialist a keyword map, the community manager the FAQ. Nobody repeats themselves, everyone picks up where they left off, and the lead base downloads as CSV.

Based on Letta, CrewAI

03

Memory without contradictions

Facts about your business and your preferences live in memory. New notes are checked against old ones: added or updated, while outdated ones move to history. The team never gets confused and loses nothing.

Based on Letta (MemGPT), mem0, Graphiti (Zep)

04

Lessons tested by results

Reject a piece and say why — the agent turns it into a rule. Approved work adds votes to the rule, and a rule that gets in the way is switched off by the team itself. Had to say it twice? The lesson becomes a hard rule.

Based on Reflexion, ExpeL, ACE

05

Skills that prove themselves

After three approvals the agent writes itself a skill to your taste. Every night it tries to improve it, but a new version stays only if it beats the old one on past tasks by a statistical test. If it starts getting rejected, it rolls back on its own. One employee's lessons become the whole team's.

Based on SkillOpt (Microsoft), SkillClaw, GEPA, Agent Workflow Memory

06

Reflection and a morning plan

Once a day the lead steps back: what you approve, what you reject, what works with your audience. And every morning it proposes the day's plan.

Based on Generative Agents

07

Task chains

Workers hand work to each other: the hunter finds companies — the content maker writes them emails; the designer makes a photo — the video maker films in the same style.

Based on LangGraph, MetaGPT

08

Deep web research

The agent searches Google, opens websites and contact pages, and finds even hidden emails and WhatsApp numbers. Every phone and email in the report is checked against the company's own site. All on one key.

Based on GPT Researcher, node-DeepResearch, crawl4ai