Examples

Cost-controlled agent

Three middleware to reduce cost on a busy agent:

# axl-config/agents/assistant/agent.toml

# Serve a repeated identical model call from a stored turn.
[middleware.llm_cache]
enabled = true

# Send short, easy requests to a cheaper model.
[middleware.model_router]
enabled = true

# Reuse a repeated lookup within one conversation.
[middleware.tool_cache]
enabled  = true
ttl_secs = 300

The router runs before the LLM cache, so a cached answer stays keyed to the model that actually produced it - you will not get a cheap model's answer served for an expensive model's request.

Bound the run itself as well:

[sub_agents]
enabled = true
max_budget_usd = 1.00
timeout_secs = 300

Check what resolved:

axl weave print --config axl-config/agents/assistant/agent.toml

Next

Approval-gated agent

A person confirms before the agent does anything consequential.

Agent with an MCP server

External tools over stdio and HTTP.

On this page

Next