Go further
Best practices
Store useful decisions, write better prompts, and keep your memory relevant to the work.
In this guide
#How a Session Should Flow
When configured properly, Claude uses Recall automatically throughout your session. Here's the ideal flow:
You start a session
Claude loads your project context automatically — past decisions, active rules, coding patterns, pending to-dos, and any in-progress workflows.
You work and make decisions
When you make a technical decision (e.g., choosing JWT over sessions), Claude saves the decision along with its reasoning so it's remembered next time.
You hit a large file or document
If the content is too large for Claude's context window, it automatically uses the RLM pipeline to process it in chunks rather than failing.
Important discussion happens
Claude extracts key insights from the conversation — requirements, patterns, bug fixes — and stores them for future sessions.
You wrap up for the day
Claude creates a session summary capturing what was accomplished. Next session starts with all of this context already loaded.
What Changes with Recall?
The key difference is that you stop manually managing context.
| Without Recall | With Recall |
|---|---|
| "Let me re-explain the project..." | Context loaded automatically at session start |
| "Remember, we decided to use JWT..." | Decisions saved and recalled automatically |
| "This file is too big to paste..." | Large content processed in chunks via RLM |
| "Where did we leave off last time?" | Session summaries picked up automatically |
Context Types
Use the appropriate type when storing memories
directiveRules Claude must followdecisionChoices made with reasoningcode_patternEstablished code conventionspreferenceUser preferences and stylerequirementProject requirementsinformationGeneral contextinsightLearned observationserrorBug patterns to avoidtodoTasks to rememberheadingSection markersImportance Scale
Low (Transient)
Temporary context, can be forgotten
Medium (General)
Useful context, worth keeping
High (Critical)
Auto-indexed, always loaded at session start
Claude Code Prompt Cookbook
Add these to your project's CLAUDE.md file to make Claude use Recall proactively without being asked
CLAUDE.md Snippet — Full Recall Integration
Add this block to your project's CLAUDE.md file. It teaches Claude how and when to use every Recall feature:
## Memory Management
You have automatic permission to use ALL Recall MCP commands
(mcp__recall-remote__*) without asking.
### Session Start
- Always call set_workspace with the project path first
- Then call auto_session_start to load context
- Check for active workflows with get_active_workflow
### During Work
- Call quick_store_decision after architectural decisions
- Store important patterns, requirements, and bug fixes
- If working on a multi-session task, use start_workflow
### End of Session
- Pause any active workflow before ending
- Call summarize_session for significant work sessions
- Run auto_consolidate periodically to keep memories clean
Quick-Reference Prompts
Paste any of these directly into Claude Code:
Set up the workspace for this project, load my context, and check for any active workflows.
The all-in-one session start. Claude calls set_workspace → auto_session_start → get_active_workflow in sequence.
Search my memories for anything related to the database schema decisions we've made.
Claude calls recall_relevant_context or search_memories to find decision-type memories about database schemas.
Store a directive: Always use snake_case for database columns and camelCase for TypeScript properties in this project.
Claude calls store_memory with context_type 'directive' and importance 9. This gets loaded at every session start.
We're going to work on the checkout refactor over the next few days. Start a workflow for it.
Claude calls start_workflow. All subsequent memories in this and future sessions are auto-tagged to the workflow.