Skip to Content
MarshalGetting started

Getting started with Marshal

Stand up Marshal so production agents share memory, skills, prompts, tools, and gateway routing under one org control plane.

Path

Sign in and create an API key

  1. Open console.exemplar.dev  and sign in (create or join an org).
  2. Create an org platform key under Tokens and API keys (eis_*).
  3. Export it for local use:
export EXEMPLAR_API_KEY=eis_…

Org keys power Marshal SDK, CLI, Gateway, and Tools & MCP. (Relay IDE hooks need a personal Relay-scoped key—see SDKs & CLI.)

Connect your estate

Enable DevOps and cloud connectors so Context Lake and tools stay current. Console: Connectors · Catalog: exemplar.dev/integrations  · Docs: Context management.

Compose prompts, skills, and memory

In the console (Marshal surfaces) or via CLI/SDK:

Route LLM traffic through the AI Gateway

Point OpenAI-compatible clients at production:

export EXEMPLAR_GATEWAY_URL=https://production-api.exemplar.dev/gateway/v1

Details: AI Gateway getting started. Add guardrails (budgets, breakers) when you promote agents.

Hello world with the SDK

pip install exemplar-harness-sdk # or: npm install @exemplar-dev/exemplar-harness-typescript-sdk
from exemplar_harness import Harness harness = Harness.from_env() memory = harness.memory(user_id="you", session_id="hello") memory.add("Prefers short answers.", memory_type="preference") print(memory.recall("how should I answer?", format="markdown"))

More: Marshal SDK · Live SDK examples · CLI · Tools & MCP.

Operate

Ingest agent sessions, then run evalsInsights (judge) and/or Session Evals — from the SDK (auto_judge_run / auto_session_eval) or the console before promoting prompt or skill changes. Pair IDE coding-agent controls with Relay when Cursor or Claude sessions need bash, path, MCP, and prompt policy.

Next

Last updated on