Live SDK examples

These demos hit the live Exemplar API when EXEMPLAR_API_KEY is set. Prefer them over simulated CI stubs when teaching agents or verifying an integration.

Home for customer samples: Exemplar-Dev/exemplar-platform-samples — Python and TypeScript in one repo, using the published SDKs.

Path in samples repo	SDK package
`python/`	`exemplar-harness-sdk`
`typescript/`	`@exemplar-dev/exemplar-harness-typescript-sdk`

Maintainer-only full matrices also remain under each SDK repo (examples/live, examples/relay, examples/ingest).

Set EXEMPLAR_API_KEY (org eis_* key). Do not use HARNESS_EXAMPLES_SIMULATED=1 for real platform traffic.

Run (Python)


git clone https://github.com/Exemplar-Dev/exemplar-platform-samples.git
cd exemplar-platform-samples/python
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env   # set EXEMPLAR_API_KEY
 
python -m platform.memory
python -m platform.skills
python -m platform.prompts
python -m platform.hitl
 
python -m relay.evaluate
python -m relay.langchain
python -m relay.agno
# more: python/relay/README.md
 
pip install -r requirements-frameworks.txt
GOOGLE_API_KEY=... python -m frameworks.langchain_mcp

Run (TypeScript)


git clone https://github.com/Exemplar-Dev/exemplar-platform-samples.git
cd exemplar-platform-samples/typescript
npm install
cp .env.example .env   # set EXEMPLAR_API_KEY
 
npx tsx relay/evaluate.ts
npx tsx relay/openai-agents.ts
npx tsx relay/claude-agent.ts
npx tsx relay/mastra.ts
npx tsx ingest/openai.ts

Platform features

Feature	Docs	Samples repo
Memory	Memory	`python/platform/memory.py`
Skills	Skill management	`python/platform/skills.py`
Prompts	Prompt management	`python/platform/prompts.py`
HITL	HITL	`python/platform/hitl.py`

Tools & MCP + frameworks

Framework	Python sample	TypeScript
LangChain	`frameworks/langchain_mcp.py`	`ingest/langchain.ts`
Agno	`frameworks/agno_mcp.py`	—
Google ADK	`frameworks/google_adk_mcp.py`	—
Claude Agent SDK	`frameworks/claude_agent_mcp.py`	`ingest/claude-agent.ts`
OpenAI / Anthropic / Vercel AI	—	`ingest/`

Agent hooks that ingest (and can queue Insights / Session Evals): Evals → Agent hooks · Client usage.

Docs: Frameworks with Marshal SDK · Tools & MCP · Evals

Relay policy adapters (in-process)

Full folders: python/relay/ · typescript/relay/

Surface	Python	TypeScript
Hook-free evaluate	`evaluate.py`	`evaluate.ts`
LangChain	`langchain.py`	`langchain.ts`
LangGraph	`langgraph.py`	(LangChain.js middleware)
Agno	`agno.py`	—
Google ADK	`adk.py`	—
Claude Agent SDK	`claude_sdk.py`	`claude-agent.ts`
OpenAI Agents	`openai_agents.py`	`openai-agents.ts`
Pydantic AI	`pydantic_ai.py`	—
CrewAI	`crewai.py`	—
Semantic Kernel	`semantic_kernel.py`	—
Mastra	—	`mastra.ts`

IDE hooks (Cursor / Claude / Codex / OpenCode) still use Relay Connect / exemplar-skills — not these adapters.

Samples repo: exemplar-platform-samples
Client usage — copy-paste API patterns
SDKs & CLI — when to use UI vs SDK vs CLI
Install IDE agent skills — /exemplar-harness and /exemplar-relay