Memory Resource

Note: This capability is now provided as an installable component. See the Components guide for how to install and use it.

Install: kdeps component install memory

Usage: run: { component: { name: memory, with: { action: "store", key: "...", value: "..." } } }

The Memory component provides persistent key-value storage across invocations, backed by a local SQLite database.

Note: The component supports store and retrieve operations. For semantic/vector memory (cosine similarity recall), use the Embedding component with a SQL resource for storage and retrieval.

Component Inputs

Input	Type	Required	Default	Description
action	string	no	store	Operation: store or retrieve
key	string	yes	—	Key identifier for the memory entry
value	string	no	—	Value to store (required when action: store)
dbPath	string	no	~/.kdeps/memory.db	Path to the SQLite database file

Using the Memory Component

Store a fact:

run:
  component:
    name: memory
    with:
      action: store
      key: "user-preference-theme"
      value: "dark"

Retrieve a fact:

run:
  component:
    name: memory
    with:
      action: retrieve
      key: "user-preference-theme"

Access the result via output('<callerActionId>').

---

Result Map

Store:

Key	Type	Description
success	bool	true when the value was stored.
key	string	The key that was stored.

Retrieve:

Key	Type	Description
success	bool	true when the key was found.
key	string	The key that was retrieved.
value	string	The stored value, or empty if not found.

---

Expression Support

All fields support KDeps expressions:

run:
  component:
    name: memory
    with:
      action: store
      key: "session-{{ get('session_id') }}-preference"
      value: "{{ get('user_preference') }}"
      dbPath: /var/lib/kdeps/memory.db

---

Full Example: Agent with Persistent Memory

A two-endpoint agent: /learn stores facts, /ask retrieves context and answers using an LLM.

# workflow.yaml
settings:
  name: memory-agent
  targetActionId: respond
  apiServerMode: true
  apiServer:
    routes:
      - path: /learn
        methods: [POST]
      - path: /ask
        methods: [POST]

# resources/learn.yaml
apiVersion: kdeps.io/v1
kind: Resource

metadata:
  actionId: storeFact
  name: Store Fact

run:
  validations:
    routes: [/learn]
    methods: [POST]

  component:
    name: memory
    with:
      action: store
      key: "fact-{{ info('timestamp') }}"
      value: "{{ get('body').fact }}"
      dbPath: /var/lib/kdeps/memory/agent.db

  apiResponse:
    success: true
    response:
      stored: true
      key: "{{ output('storeFact').key }}"

# resources/recall.yaml
apiVersion: kdeps.io/v1
kind: Resource

metadata:
  actionId: recallFact
  name: Recall Fact

run:
  validations:
    routes: [/ask]
    methods: [POST]

  component:
    name: memory
    with:
      action: retrieve
      key: "{{ get('body').key }}"
      dbPath: /var/lib/kdeps/memory/agent.db

# resources/respond.yaml
apiVersion: kdeps.io/v1
kind: Resource

metadata:
  actionId: respond
  name: Answer Question
  requires: [recallFact]

run:
  validations:
    routes: [/ask]
    methods: [POST]

  chat:
    model: llama3
    prompt: |
      Answer using the context below. If context is empty, say you don't know.

      Context: {{ output('recallFact').value }}

      Question: {{ get('body').question }}

  apiResponse:
    success: true
    response:
      answer: "{{ output('respond') }}"

Run:

kdeps run workflow.yaml

# Store a fact
curl -X POST http://localhost:16394/learn \
  -H 'Content-Type: application/json' \
  -d '{"fact": "The capital of France is Paris."}'

# Ask a question
curl -X POST http://localhost:16394/ask \
  -H 'Content-Type: application/json' \
  -d '{"key": "fact-...", "question": "What is the capital of France?"}'