Agent development¶
Your project supplies the Agent's behavior. Market Lab supplies experiment time, market access, simulated trading, recording and process management. You do not construct those components yourself.
Write create_agent(context)
→ import a file or folder to save a revision
→ prepare its Python and dependencies
→ select that revision in an experiment
→ Play creates work for the prepared Agent
| Step | Guide |
|---|---|
| Write the factory, loop and local tools | SDK and Agent interface |
| Save source, configure capabilities and prepare dependencies | Agent projects |
| Select or provision a compatible Python | Python selection |
| Start from a published architecture (TradingAgents, FinMem, FinCon) | Reference Agents |
Use the built-in template to start quickly, or return your own run(task) / run() object.
Examples live in examples/agents.
Platform tool proxies come from context; your own tools run inside the Agent process.
For the underlying communication route, see processes and tools.
Maintained source: docs/agent-development/README.md