Skip to main content

​
Problem

You need to process multiple related tasks where later jobs should have access to the results of earlier ones. Each job should build on accumulated knowledge.

​
Solution

Store job results in a persistent context file that grows over time. Each new job receives the entire accumulated memory as context.

​
Example

Build knowledge across research tasks: 
# Save results to memory
flw ./workspace $job1 >> memory.txt
flw ./workspace $job2 >> memory.txt
flw ./workspace $job3 >> memory.txt

# Use memory as context
wrk ./workspace "Given this context:
$(cat memory.txt)

Answer: What's the best approach for a startup?"

​
Agent Loop with Memory

Continuous iteration with accumulated context: 
GOAL="Write a Python tutorial covering variables, loops, and functions"
memory=""

for i in {1..5}; do
  result=$(wrk ./workspace "Goal: $GOAL
Previous work: $memory
Continue. Write the next section. Say DONE if complete." | xargs flw ./workspace)

  echo "=== Iteration $i ==="
  echo "$result"

  if echo "$result" | grep -q "DONE"; then
    break
  fi

  memory="$memory\n---\n$result"
done

​
How It Works

  • Results are appended to a file (memory.txt)
  • The entire file is read and included in subsequent prompts
  • The model sees the full history and can reference previous work
  • The loop continues until a termination condition is met

​
Memory Management

Structured memory: 
# JSON format for structured data
echo "{\"task\": \"research\", \"result\": \"$result\"}" >> memory.jsonl

# Use jq to query
cat memory.jsonl | jq -s '.' | wrk ./workspace "Analyze this data: $(cat -)"
Selective memory: 
# Only keep the last N results
tail -n 100 memory.txt > memory.tmp
mv memory.tmp memory.txt
Categorized memory: 
# Different memory files for different topics
flw ./workspace $db_research >> memory_database.txt
flw ./workspace $api_research >> memory_api.txt

# Use relevant context only
wrk ./workspace "Context: $(cat memory_database.txt)\nQuestion: Best DB for this use case?"

​
Use Cases

  • Multi-step research: Each query builds on previous findings
  • Iterative writing: Continue long-form content across multiple jobs
  • Session context: Maintain conversation state across interactions
  • Progressive refinement: Gradually improve output based on accumulated feedback

Build docs developers (and LLMs) love

Get started for freeTalk to us