examples

TokenJam Examples

Example agents demonstrating tj integrations, from single-provider basics to complex multi-agent workflows.

Quick Start

pip install -e ".[dev]"           # install tokenjam in dev mode
export ANTHROPIC_API_KEY=sk-...   # set your API key
python examples/single_provider/anthropic_agent.py

After running any example, inspect the results with:

tj status       # agent session overview
tj traces       # list all spans from the run
tj cost --since 1h   # cost breakdown
tj alerts       # any fired alerts

---

Single Provider

One integration per file. Simplest way to see tj in action with your provider of choice.

Example	Env Vars	Extra Deps	Description
anthropic_agent.py	ANTHROPIC_API_KEY	anthropic	Tool-use agent with calculator and weather tools
openai_agent.py	OPENAI_API_KEY	openai	Function-calling agent with streaming response
gemini_agent.py	GOOGLE_API_KEY or GEMINI_API_KEY	google-generativeai	Text summarization via Gemini Flash
bedrock_agent.py	AWS_DEFAULT_REGION + AWS creds	boto3	Claude on AWS Bedrock (advanced setup)
openai_agents_sdk_agent.py	OPENAI_API_KEY	openai-agents httpx	Multi-agent handoff via OpenAI Agents SDK
litellm_agent.py	OPENAI_API_KEY ANTHROPIC_API_KEY	litellm	Multi-provider routing via LiteLLM

Note: openai_agents_sdk_agent.py requires tj serve running (tj serve &).

---

Single Framework

One framework integration per file. Shows how tj captures framework-level spans.

Example	Env Vars	Extra Deps	Description
langchain_agent.py	OPENAI_API_KEY	langchain-core langchain-openai	Tool-calling agent with calculator and word counter
langgraph_agent.py	OPENAI_API_KEY	langgraph langchain-openai	Plan-execute-review graph pipeline
crewai_agent.py	OPENAI_API_KEY	crewai	Researcher + writer crew collaboration
autogen_agent.py	OPENAI_API_KEY	pyautogen	Two-agent debate with back-and-forth
llamaindex_agent.py	OPENAI_API_KEY	llama-index	RAG query engine over sample documents

Note: llamaindex_agent.py requires tj serve running (tj serve &).

---

Multi-Integration

Complex real-world patterns combining multiple providers and frameworks. These showcase tj's ability to track cost, performance, and behavior across a heterogeneous agent stack.

Example	Env Vars	Extra Deps	Description
router_agent.py	ANTHROPIC_API_KEY OPENAI_API_KEY GOOGLE_API_KEY	anthropic openai google-generativeai	Routes tasks to the cheapest/best provider
research_team.py	ANTHROPIC_API_KEY OPENAI_API_KEY	anthropic crewai langchain-core	CrewAI agents with LangChain tools
rag_pipeline.py	OPENAI_API_KEY ANTHROPIC_API_KEY	llama-index openai anthropic	RAG with OpenAI-to-Anthropic fallback

Note: rag_pipeline.py requires tj serve running (tj serve &).

---

Alerts and Drift

These examples demonstrate what makes tj unique: real-time alerting and behavioral drift detection. No API keys required -- they use simulated instrumentation via record_llm_call() and record_tool_call().

Example	Env Vars	Description
sensitive_actions_demo.py	None	Fires alerts when agent calls sensitive tools
budget_breach_demo.py	None	Exceeds budget limits, shows cost alerts
drift_demo.py	None	Builds baseline, then triggers drift detection

These examples include the required tj.toml config snippets as comments at the top of each file. Copy the relevant config to your tj.toml before running.

# No API keys needed -- just run:
python examples/alerts_and_drift/budget_breach_demo.py

# Then inspect:
tj alerts       # see budget-breach alerts
tj cost --since 1h   # see cost tracking

---

Which examples need tj serve?

Most examples use in-process telemetry (spans go directly to the local DuckDB). These two integrations export via OTLP HTTP and require the server:

• openai_agents_sdk_agent.py
• llamaindex_agent.py
• rag_pipeline.py (uses LlamaIndex)

Start the server before running them:

tj serve &