crewai-soul 0.2.0

Markdown-native memory for CrewAI agents. Human-readable, git-versionable, no database required.

crewai-soul 🧠

Markdown-native memory for CrewAI agents.

Your crew's memory shouldn't require a database. crewai-soul stores everything in two simple markdown files:

• SOUL.md — Agent identity (who it is, how it behaves)
• MEMORY.md — Timestamped log of all interactions

Human-readable. Git-versionable. No infrastructure required.

Install

pip install crewai-soul

Quick Start

from crewai import Crew, Agent, Task
from crewai_soul import SoulMemory

# Create markdown-based memory
memory = SoulMemory()

# Use it with your crew
crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, writing_task],
    memory=memory,
)

result = crew.kickoff()

After running, check MEMORY.md:

# Memory Log

## 2026-03-06 22:30:15 UTC
The researcher found that PostgreSQL handles 10k concurrent connections.

## 2026-03-06 22:31:02 UTC
The writer produced a technical comparison document.

Why Markdown?

Feature	CrewAI Built-in	crewai-soul
Storage	Vector database	Markdown files
Human-readable	❌	✅
Git-versionable	❌	✅
Needs LLM to store	Yes (scope inference)	No
Infrastructure	Database server	None
Audit trail	Complex	Just read the file

API

SoulMemory

from crewai_soul import SoulMemory

memory = SoulMemory(
    soul_path="SOUL.md",      # Agent identity
    memory_path="MEMORY.md",  # Memory log
)

# Store a memory
memory.remember("We decided to use PostgreSQL.")

# Recall relevant memories
matches = memory.recall("What database did we choose?")
for m in matches:
    print(f"[{m.score:.2f}] {m.content}")

# Clear memories
memory.forget()

# Get stats
print(memory.info())

With Scopes

# Store with a scope tag
memory.remember("Sprint velocity is 42 points", scope="/team/metrics")

# Recall from specific scope
matches = memory.recall("velocity", scope="/team")

Per-Agent Memory

from crewai import Agent
from crewai_soul import SoulMemory

# Each agent gets its own memory file
researcher = Agent(
    role="Researcher",
    goal="Find information",
    memory=SoulMemory(memory_path="researcher_memory.md"),
)

writer = Agent(
    role="Writer", 
    goal="Write reports",
    memory=SoulMemory(memory_path="writer_memory.md"),
)

Standalone Usage

Works without CrewAI too:

from crewai_soul import SoulMemory

memory = SoulMemory()

# Build up knowledge
memory.remember("The API rate limit is 1000 requests per minute.")
memory.remember("Our staging environment uses port 8080.")

# Later, recall what you need
matches = memory.recall("What are our API limits?")

Upgrading to Semantic Search

For better recall on large memory files, install with RAG support:

pip install crewai-soul[rag]

This uses soul-agent's hybrid RAG+RLM retrieval under the hood.

Links

• soul.py — The core library
• CrewAI — Multi-agent framework
• SoulMate API — Hosted memory service

License

MIT