# Flowlines

> Behavioral observability and structured memory for production AI agents. Flowlines detects when your agents drift, lose context, or frustrate users across sessions, then surfaces the exact memory fix that would prevent the same failure next time. You approve the fix, Flowlines applies it.

Flowlines sits at the layer between execution observability (LangSmith, Langfuse, Arize) and memory stores (Mem0, Zep, Letta). It works alongside both on day one and replaces them as you grow.

## What Flowlines does

- **Traces.** Two-line SDK install captures every agent call, prompt, tool use, and decision as a structured execution graph.
- **Behavioral signals.** Every trace is scored for failure modes: agent drift, context loss, user frustration, repetition, constraint violations. When a pattern correlates with a missing memory field, Flowlines surfaces the exact fix with statistical evidence.
- **Structured memory.** Once a fix is approved, Flowlines extracts typed, scoped, versioned memory from traces (preferences, constraints, task state, recurring patterns) and injects it into future calls. Reversible.

## How Flowlines is different

- **Versus execution observability (LangSmith, Langfuse, Arize):** those tools show what happened inside one LLM call. Flowlines shows how an agent is behaving across sessions and users, and what to fix.
- **Versus memory stores (Mem0, Zep, Letta):** those tools give your agent a place to write things down. Flowlines tells you which fields are missing, why they matter, and surfaces the evidence.
- **The wedge:** detection across sessions is how Flowlines lands. Structured memory is how it stays.

## Use cases

- [Coding agents](https://flowlines.ai/use-cases/coding-agents): catches recurring vulnerability patterns (SQL injection, hardcoded secrets), drafts fixes based on developer history, compresses acceptance time with every occurrence.
- [Customer support](https://flowlines.ai/use-cases/customer-support): detects repeat contacts, declining sentiment, and resolution failures; surfaces the missing context fields before the third escalation.
- [EdTech and tutoring](https://flowlines.ai/use-cases/edtech): detects student frustration through language patterns and engagement drops, correlates with missing avoidance_patterns, surfaces the field that would reduce repeat frustration.
- [Healthcare (preview)](https://flowlines.ai/use-cases/healthcare): typed memory fields with full provenance for regulated environments, every write traceable to the interaction that produced it.

## Integration

- Two-line SDK install (Node.js, Python)
- Works with any LLM provider (OpenAI, Anthropic, open-source)
- Works with any agent framework
- Free during early access

## Links

- Website: https://flowlines.ai
- How it works: https://flowlines.ai/#demo
- All use cases: https://flowlines.ai/use-cases
- Blog: https://flowlines.ai/blog
- Request early access: https://flowlines.ai/request-access

## Contact

- Email: alex@flowlines.ai
- Founder: Alexandre Ayoub
- Company: Flowlines SAS, Paris, France