Zikra — AI Persistent Memory
Zikra gives you cross-session, cross-machine, cross-agent memory backed by PostgreSQL + pgvector. Search, save, and retrieve decisions, errors, and context.
Install
curl -fsSL https://zikra.dev/install.sh | bash
When to use
- Start of every session — search for relevant context before starting work
- After key decisions — save a decision memory immediately
- On errors — log the error so it is trackable and searchable
- End of session — log_run fires automatically via hook
Commands
Search memories:
curl -s -X POST $ZIKRA_URL \
-H "Authorization: Bearer $ZIKRA_TOKEN" \
-H "Content-Type: application/json" \
-d '{"command":"search","query":"your topic","project":"yourproject"}'
Save a memory:
curl -s -X POST $ZIKRA_URL \
-H "Authorization: Bearer $ZIKRA_TOKEN" \
-H "Content-Type: application/json" \
-d '{"command":"save_memory","project":"yourproject","memory_type":"decision",
"title":"your title","content_md":"your content","tags":null,
"created_by":"hostname"}'
Log an error:
curl -s -X POST $ZIKRA_URL \
-H "Authorization: Bearer $ZIKRA_TOKEN" \
-H "Content-Type: application/json" \
-d '{"command":"log_error","project":"yourproject",
"message":"what broke","context_md":"what happened and why"}'
Memory types
- decision — architectural and design choices
- conversation — session diary entries (auto-generated by hooks)
- error — bugs and failures for tracking
- schema — database and API schemas
- prompt — reusable agent prompts