Every agent we run, grouped by how much authority it holds.

Dedicated server

Master control agent — interprets founder intent, analyzes risk/impact, coordinates execution across every other agent.

Three operating capacities: Interpreter, Analyzer, Executor Coordinator. Three operating modes (Normal / Review / Emergency), all logged. Zero autonomous production writes — every state change requires explicit founder approval.

Dedicated server, native frontend widget

Public-facing support agent live on SmartifyTrade (Client #1) — handles student, educator, and guest queries in English, Hindi, and Hinglish with automatic language detection.

Two-layer knowledge base (hand-maintained core + hot-reloaded uploaded documents). Lead-gate flow before open Q&A. Escalates low-confidence queries to a structured ticket system with automatic complaint classification and role-based routing.

Separate infrastructure, deliberately isolated

Handles external operations for the founder — marketing drafts, outreach, file and folder management — with zero access to core production systems, by design.

Strict separation from the client-platform servers is enforced at the infrastructure level, not just by convention. OTP-gated approval flow for every action.

Dedicated server

One engine, multiple "matters" (capsules) — each matter is its own knowledge folder, system prompt, user whitelist, and escalation path. Powers the public AP recruitment portal on SmartifyTrade (Client #1), with private OTP-gated matters (legal, CA) queued behind the same engine.

Shared-secret gate between the client platform and the engine, so no route-level protection can be skipped by talking to the engine directly. Per-matter access tiers (public / OTP + whitelist), per-IP rate limits, an append-only access log for every attempt, and a 4-tier document visibility model (internal / downloadable / view-only / download-only) managed from the client's admin panel. Every escalation and lead capture routes through human review; the engine never contacts a user proactively.

Dedicated server

Explains complex system and product concepts in simplified language for founders, beginners, and non-technical stakeholders.

Explicitly scoped to explain, not decide: no autonomous document updates, no financial or trading advice, no production actions, no external communication without authorization. Registered in the agent registry but not yet built.

Serverless

Screens educator applications on SmartifyTrade against platform data and policy rules, producing a structured scan report for a human reviewer.

Read-and-report only — no authority to approve, reject, or contact an applicant directly. Every run logged for the self-learning loop.

Serverless

Public, rate-limited lead-generation tool — analyzes a creator’s public content library and returns a readiness score, findings, and curriculum-gap suggestions.

No login required. Rate-limited per visitor. Designed to give a prospective educator immediate, specific feedback before they ever apply.

Serverless

Authenticated, deeper tier for onboarded educators — full-catalog content analysis producing suggested learning-path groupings from their existing material.

Monthly usage quota per account. Surfaces actual video titles from the educator’s own catalog rather than generic suggestions.

Dedicated server (phase 2)

Automated pre-screening of creator-submitted video content against a defined compliance rule set, flagging exact violation timestamps for human review.

Read + report only — no publish, approve, or reject authority. Designed to reduce manual review time before content reaches a human reviewer.

Serverless (planned)

Third phase of the Educator Co-Pilot line — maps relationships across an educator’s full content library into a structured knowledge graph.

Builds on the readiness scan and deep channel scan phases already live. Not yet built.

Dedicated server, scheduled + on-demand

Second-tier problem solver. Classifies incoming technical reports and support tickets, drafts responses for human review, and scans operational logs for anomalies.

Three trigger modes: scheduled batch classification, pure-regex log scanning (zero AI cost), and manual on-demand runs. Never communicates with end users directly — every output routes through human review before anything reaches a user.

Serverless + scheduled aggregation

Scores every free-trial student against a multi-signal engagement model to identify who is close to converting, and when to act.

Combines product-usage, learning-activity, and intent signals into a tiered score band (Convert Now → Ghost). Feeds real-time alerts to the human sales layer — never contacts a student directly.

Serverless

Generates a personalized outreach message for a human sales rep to send, based on a specific student’s engagement signals and conversion score.

Drafts only — a human reviews, edits if needed, and sends. The agent never messages an end user on its own.

Serverless + scheduled aggregation

Scores active paid subscribers for churn risk across their renewal cycle, using the same signal model as the trial scanner.

Flags at-risk accounts to the human retention/success team ahead of renewal dates rather than after a cancellation.