| name | akasha-memory |
| description | Store, retrieve, and query shared memory records in Akasha — a distributed cognitive fabric
for intelligent agent collaboration. Use when the agent needs persistent memory, wants to
share findings with other agents, or needs to read context written by other agents.
Supports working, episodic, semantic, and procedural memory layers with automatic
consolidation via the Nidra engine.
|
| license | ASL-1.0 |
| compatibility | Requires curl or httpx. Works with any LLM agent that can make HTTP requests. |
| metadata | {"author":"ocuil","version":"1.0.4","category":"memory","tags":"memory shared-memory agents collaboration distributed"} |
Akasha Memory Skill
You have access to Akasha, a shared cognitive fabric. Use it to persist and retrieve
structured memory across sessions and across agents.
Configuration
Before using this skill, ensure you have:
AKASHA_URL: The Akasha endpoint (e.g., https://localhost:7777)
AKASHA_TOKEN: A valid JWT token or API key (e.g., ak_xxx)
Memory Layers
| Layer | Path prefix | Purpose | Lifecycle |
|---|
| Working | memory/working/ | Active task state, scratchpad | Auto-expires (TTL) |
| Episodic | memory/episodic/ | Events, experiences, interactions | Consolidates via Nidra |
| Semantic | memory/semantic/ | Distilled knowledge, insights | Permanent |
| Procedural | memory/procedural/ | How-to instructions, recipes | Permanent |
Operations
Write a record
curl -sk -X POST "$AKASHA_URL/api/v1/records/<namespace>/<path>" \
-H "Authorization: Bearer $AKASHA_TOKEN" \
-H "Content-Type: application/json" \
-d '{"value": {"key": "value"}, "ttl_secs": 3600}'
ttl_secs is optional. Omit for permanent records.
value can be any valid JSON object.
Read a record
curl -sk "$AKASHA_URL/api/v1/records/<namespace>/<path>" \
-H "Authorization: Bearer $AKASHA_TOKEN"
Query records (glob pattern)
curl -sk "$AKASHA_URL/api/v1/query?pattern=memory/working/**" \
-H "Authorization: Bearer $AKASHA_TOKEN"
Delete a record
curl -sk -X DELETE "$AKASHA_URL/api/v1/records/<namespace>/<path>" \
-H "Authorization: Bearer $AKASHA_TOKEN"
Recommended Path Structure
memory/working/<agent-id>/current-task → Active task context
memory/working/<agent-id>/scratchpad → Temporary notes
memory/episodic/<domain>/<event-id> → What happened
memory/semantic/<domain>/<insight-id> → What was learned
memory/procedural/<domain>/<procedure-id> → How to do things
Best Practices
- Always namespace by agent-id for working memory to avoid collisions
- Use semantic memory for important conclusions you want other agents to find
- Set TTLs on working memory — it's meant to be ephemeral
- Query before writing — check if similar knowledge already exists
- Use descriptive paths — they serve as the primary discovery mechanism
Python SDK
If available, use the Python SDK for a cleaner API. See references/python-sdk.md.