Features & Use Cases¶
Discover how autonomous AI agents are transforming industries and workflows.
MiMinions empowers developers to create intelligent, autonomous agents that handle complex tasks, maintain context, and integrate seamlessly with external tools. Explore real-world applications below.
Knowledge-Aware AI Agents¶
Memory & Context
Build agents with long-term memory using vector storage. Your agents can store, recall, and update knowledge dynamically, maintaining context across conversations.
- Store and retrieve knowledge efficiently
- Semantic search with similarity scoring
- Update and manage memory entries
- Context-aware responses
Attach a memory backend when you create the agent — store_knowledge() writes
straight into it.
from miminions.memory.sqlite import SQLiteMemory
from miminions.agent import create_minion
agent = create_minion("Assistant", memory=SQLiteMemory("agent.db"))
agent.store_knowledge(
"Python is a high-level programming language",
metadata={"category": "programming"},
)
results = agent.recall_knowledge("What language is Python?")
Perfect for
- Customer support agents that remember user history
- Personal assistants with long-term context
- Research agents that accumulate knowledge
- Educational tutors with personalised learning
Memory is required
store_knowledge() and recall_knowledge() raise ValueError if no memory
is attached. The [sqlite] extra (pip install miminions[sqlite]) installs
the vector backend. See Memory for the full CRUD and
search API.
Intelligent Document Processing¶
Document Intelligence
Ingest and process documents (PDFs, text, and Markdown) with automatic chunking for optimal retrieval. Agents can understand and query document content with semantic search.
- PDF, text, and Markdown ingestion
- Automatic text chunking with overlap
- Semantic document search
- Metadata tagging and filtering
With memory attached, the built-in ingest_document tool chunks the file and
stores every chunk in vector memory.
from miminions.memory.sqlite import SQLiteMemory
from miminions.agent import create_minion
agent = create_minion("DocAgent", memory=SQLiteMemory("docs.db"))
result = agent.execute_tool("ingest_document", filepath="resume.pdf")
print(result["chunks_stored"])
# Now query the document semantically
matches = agent.recall_knowledge("years of Python experience")
Perfect for
- Legal document analysis and Q&A
- Resume screening and candidate matching
- Research paper summarisation
- Contract review and compliance
Requirements
ingest_document needs a memory backend attached. Supported file types are
.pdf, .txt, .md, and .text — other extensions raise ValueError. PDF
extraction requires the optional pdfplumber package
(pip install pdfplumber).
Model Context Protocol (MCP) Integration¶
Tool Integration
Connect your agents to external tools and services using the Model Context Protocol. Dynamically load and execute tools from MCP servers alongside custom Python functions.
- Connect to MCP servers
- Load tools dynamically
- Mix MCP and local functions
- Async tool execution
Connect to a server, then load its tools into the agent's registry.
from mcp import StdioServerParameters
from miminions.agent import create_minion
agent = create_minion("MathAgent")
await agent.connect_mcp_server(
"math",
StdioServerParameters(command="python", args=["math_server.py"]),
)
await agent.load_tools_from_mcp_server("math")
# MCP tools are now registered alongside any local functions
print(agent.list_tools())
await agent.cleanup()
Perfect for
- Workflow automation with external APIs
- Data processing pipelines
- Multi-tool agent orchestration
- Integration with existing systems
MCP package
MCP support requires the optional mcp package. See
Agent for the full connect/load lifecycle.
Persistent Knowledge with SQLite Memory¶
Data Persistence
Use SQLite-backed memory for permanent knowledge storage that persists across sessions. Perfect for agents that need to maintain state over time.
- Persistent storage across sessions
- CRUD operations on memory
- Metadata, keyword, regex, and hybrid search
- Custom database locations
Point SQLiteMemory at a file path and the knowledge survives process restarts.
from miminions.memory.sqlite import SQLiteMemory
from miminions.agent import create_minion
memory = SQLiteMemory("agent_memory.db")
agent = create_minion("PersistentAgent", memory=memory)
agent.store_knowledge("User prefers concise answers", metadata={"type": "pref"})
# Direct backend access for advanced search
hits = memory.hybrid_search("concise", top_k=3)
for hit in hits:
print(hit["text"], hit["meta"])
Perfect for
- Multi-session conversations
- Knowledge base management
- User preference tracking
- Audit trails and logging
Result shape
SQLiteMemory search results are dicts with the keys id, text, meta,
and (for vector reads) distance. The metadata key is meta, not
metadata. Full API in Memory.
Ready to build your own agent?¶
Start creating autonomous AI agents with memory, tool integration, and document processing capabilities.