Your AI tutor forgets every student. Flowlines fixes that.

Name: Flowlines
Author: Flowlines

Tutoring agents that can't remember learning preferences, avoid known triggers, or adapt to each student aren't tutoring. They're reciting. Flowlines gives them structured memory, behavioral signals, and feedback loops.

Request early access See the product

Why AI tutors fail in production

Students repeat their learning style every session because the agent doesn't retain preferences between conversations.

The agent drills topics the student already mastered or avoids, causing frustration that shows up as disengagement, not complaints.

No way to know which students are struggling silently. By the time a human reviews, the student has already churned.

Token costs balloon because the agent re-discovers context instead of loading structured state from previous sessions.

Structured memory for every learner

Flowlines traces every tutoring session, extracts structured memory fields (learning preferences, knowledge gaps, avoidance patterns, comprehension level), and injects them into future sessions. Signals detect frustration, context loss, and disengagement in real-time, before the student complains.

Signals Flowlines detects

Student frustration

72% correlation with missing avoidance_patterns

Detects frustration through language patterns, topic avoidance, and engagement drops. Correlates with missing memory fields to suggest fixes.

Context loss

42% reduction when session_history injected

Agent references or repeats information from previous sessions. Students say "you already told me this", a sign memory injection failed.

Disengagement

3-session lead time before churn

Shortening responses, topic drift, delayed replies. Early warning for at-risk students before they stop showing up.

Structured memory schema

Flowlines extracts and maintains these fields automatically from every trace.

Field	Scope	Description	Coverage
learning_preferences	user	Visual vs. text, pace preference, drill tolerance	78%
avoidance_patterns	user	Topics/methods the student actively resists	31%
knowledge_gaps	user	Specific concepts not yet mastered	54%
comprehension_level	user	Current assessed level per topic area	44%
current_topic	session	Active topic and sub-topic in this session	95%
quotes_studied	user	Literary quotes already analyzed (for lit tutors)	54%

Example session

lit-tutor-gcse/Kai · Recall drill dispute

Welcome back Kai! Let's pick up where we left off with Act 3 of Macbeth.

Ok what are we doing today

Let's start with a quick recall drill on Act 2 to warm up.

I don't get why we have to do these recall drills. Can't we just read the scenes?

Signal fired: student frustration. avoidance_patterns not injected

I hear you, Kai. Let me try a visual timeline approach instead. I know you respond better to that.

Memory write: avoidance_patterns → ["recall_drills"]

Measured impact

-23%

Frustration

After injecting avoidance_patterns

+34%

Retention

Students returning for 5+ sessions

-18%

Token cost

From structured memory injection

1 turn

Time to adapt

vs. 3+ turns without memory

Ready to make your edtech & tutoring agents reliable?

Request early access and get a direct line to the founder. We'll review your agent setup and have you running traces within the week.

Request early access