EXEMPLARExemplar is part of the NVIDIA Inception program.
NVIDIA InceptionExemplar is part of the Google for Startups program.
Google for StartupsMarshal · Long-term agent state & recall
Memory
Durable per-agent and per-user memory—facts, preferences, and session history agents recall across turns—separate from org-wide Context Lake knowledge
Marshal · Memory
How Memory works
Durable per-agent and per-user memory—facts, preferences, and session history agents recall across turns—separate from org-wide Context Lake knowledge
MEMORY
write · recall · scope
Sessions write lasting state (sky rails); agents recall across turns under org policy (orange rails). Separate from Context Lake.
Sessions write lasting state (sky rails); agents recall across turns under org policy (orange rails). Separate from Context Lake.
Why it matters
Built for production agents—not prompt playgrounds
How this harness capability fits the Exemplar platform: one identity, one policy layer, and one audit trail across console, MCP, and SDK.
Why Exemplar
Agents without memory re-ask the same preferences and standing instructions every session—burning tokens and frustrating operators.
Exemplar treats memory as harness infrastructure: scoped, searchable, and governed—not ad-hoc chat history stuffed into every prompt.
What Exemplar delivers
A multi-tenant memory store with Gemini embeddings and org-isolated collections—shared by console, MCP, and SDK clients.
Session-spanning recall so assistants in Cursor, Claude, and the console keep standing context without re-prompting.
How teams use it
Scope memory once (userId / agentId / sessionId); agents add and recall through harness.memory or the memory MCP tool.
Pair with Context Lake retrieve for grounding in catalog and connector knowledge—memory for state, lake for truth.
Capability checklist
Developer guide
Official documentation on docs.exemplar.dev for this capability.
Open developer guide (opens in a new tab)Get started
Put Memory to work with the rest of Marshal—or talk to us about your agent stack.
From the blog
Related posts on exemplar.dev.
- Moving from prompt and context engineering toward harness engineering
Three layers behind production agents: shaping the ask, assembling the window, and building the runtime loop. Where each discipline stops and what to invest in next.
- Knowledge Architecture for AI Coding Agents: Beyond the AGENTS.md File
How to structure what your AI coding agent knows — the docs directory, context boundaries, ownership, versioning, and the difference between a knowledge system that stays useful and one that silently rots.
- Agents, context, and guardrails on a unified platform
From code completion to production actions; Context Lake, catalog, governance, and DevX Assist/MCP for safe automation.