Lessons
When you correct the agent, it remembers. Lessons persist corrective knowledge across sessions so the agent never makes the same mistake twice.Lessons are saved feedback. They can capture corrections, style preferences, workflow preferences, or patterns you want Beakr to repeat because they worked well.
How lessons work
When the agent gets something wrong and the user corrects it, the correction is saved as a lesson.When a user saves feedback, Beakr stores it as a lesson with trigger cues that describe when it should apply. On future runs, the system checks whether any saved lessons match the current context and injects them into the agent's prompt -- before the agent starts reasoning.
- Feedback. The user tells the agent what to do differently next time, or what behavior to preserve.
- Save. The agent saves the lesson with trigger keywords and file extension patterns that identify when this lesson is relevant.
- Match. On future runs, the system matches the user's message and attached files against saved lesson triggers.
- Inject. Matching lessons are injected into the agent's context before it starts reasoning, so it applies the correction proactively.
Lessons vs. skills
| Memory type | What it means | How it is used |
|---|---|---|
| Skills | Capabilities and procedures: what the system can do. | Reusable workflows or instructions that can be intentionally invoked or loaded on demand. |
| Lessons | Feedback and preferences: what the system has learned from a user or team. | Injected automatically when relevant so future runs reflect prior corrections, style choices, or approved patterns. |
What a lesson contains
| Field | Purpose |
|---|---|
| Lesson content | What was wrong and what to do instead. The corrective knowledge itself. |
| Trigger keywords | Comma-separated keywords that match against user messages (e.g., "normalization, batch correction, combat"). |
| Trigger extensions | File extensions that match against attached files (e.g., .h5ad, .vcf). Useful for domain-specific corrections. |
| Hit count | How many times this lesson has been injected. Helps identify the most impactful corrections. |
Scope
Lessons are scoped to the individual user and organization. Your corrections are yours -- they don't affect other users' experiences. Row Level Security ensures lesson data is fully isolated.
Managing lessons
Users can review, edit, and dismiss lessons. Dismissed lessons stop being injected but are not deleted, so they can be re-enabled if needed.
Why this matters
Most AI tools have no memory of past corrections. Every session starts fresh, and the same mistakes recur. Lessons create a feedback loop: the more you use Beakr, the better it understands your preferences, methodologies, and domain-specific conventions. This is user-level personalization that compounds over time.