Your agent remembers across sessions

The problem

Every new session, your coding agent starts from zero. You re-explain the auth model, the naming conventions, the one service that's load-bearing-and-fragile, the decision you made last week and why. The agent is brilliant and amnesiac — and the re-explaining tax compounds across a team.

kyma gives it durable memory: facts you (or the agent) save once, recalled into the right prompts forever, across sessions and machines.

Setup

One binary, zero infra, then wire your agent (full options):

curl -fsSL https://raw.githubusercontent.com/shakedaskayo/kyma/main/install.sh | bash
kyma setup claude-code        # MCP — or: kyma install-plugin (automatic capture + recall)

Save what's durable

Save a fact explicitly from the CLI, or let the agent call save_memory over MCP:

kyma remember "payments-svc deploys behind the Aurora gateway; error budget is 0.1%."
kyma remember "Prefer KQL over SQL in examples." --type preference

Memory is typed (fact, preference, decision, …) and graph-aware: a memory about payments-svc links to the real service node, its repo, and the traces it emits. Wrap anything sensitive in <private>…</private> and it's stripped before it's ever embedded or stored.

Recall it — automatically

Ask for it explicitly:

kyma recall "how do we deploy payments and what's the error budget?"
# → returns the memory, scored by vector + keyword + graph

…but the point is you mostly don't. With the Claude Code plugin, hooks inject the most relevant memories into every prompt with no tool call — so the agent already knows before you ask. Recall is graph-aware hybrid retrieval (vector + keyword, fused with Reciprocal Rank Fusion, then expanded over the memory graph), so it returns the surrounding context, not just a string match.

What you'll see

A new session, cold, asked "how should I add a field to the checkout flow?" — and the agent answers with your conventions baked in: it knows checkout calls payments-svc, that you prefer KQL examples, that the error budget is tight. No prologue, no re-explaining.

Scale it

• Across machines & teammates — sync keeps your laptop and the team control plane coherent: memory written on one shows up on the others.
• Kept sharp automatically — dreaming reviews memories on a schedule, merges duplicates, supersedes contradictions (bi-temporal — nothing is hard-deleted), and wires memories to the resources they're about.
• From the agent's own files — kyma sync also ingests Claude Code's file memory (~/.claude/projects/*/memory) into queryable kyma memory.

Next

• The recall internals: Agentic Memory.
• The full loop in action: Debug a prod incident from your editor.
• How it all fits: How kyma works.