Lanes Reference

A lane is a named execution path that SwitchBoard selects based on task attributes. Each lane has a canonical backend that performs the actual work.

Lane inventory

Lane	Backend	Cost	Capability
aider_local	aider_local	Free	CPU-only Aider via local Ollama
claude_cli	kodo	Medium	Claude-powered Kodo executor
claude_cli	archon_then_kodo	High	Archon workflow + Kodo execution
codex_cli	kodo	Medium	Codex routing, Kodo execution

aider_local

When selected: Low-risk, bounded tasks eligible for local execution.

Default routing rules (from switchboard/lane/defaults.py):

Priority	Rule	Condition	Backend
10	force_local_only	local_only label	aider_local
20	local_low_risk	task_type ∈ {lint_fix, documentation, simple_edit} AND max_risk_level=low	aider_local
60	local_catchall	task_type ∈ {lint_fix, documentation, simple_edit}	aider_local

Fallback: If Ollama is unavailable, alternatives include claude_cli + kodo (see local_to_remote_fallback in defaults).

Infrastructure: Requires a running Ollama instance at http://localhost:11434 with qwen2.5-coder:3b pulled. See PlatformDeployment docs.

claude_cli

When selected: Medium-to-high complexity tasks, or anything not matched by a local rule.

Default routing rules:

Priority	Rule	Condition	Backend
30	medium_implementation	task_type ∈ {bug_fix, test_write, dependency_update} AND risk ∈ {low, medium}	kodo
40	premium_structured	task_type ∈ {refactor, feature} AND risk ∈ {medium, high}	archon_then_kodo
50	high_risk_escalation	risk_level=high	kodo

Fallback (global): claude_cli + kodo — used when no rule matches.

Routing policy

Policies are loaded from YAML and evaluated in ascending priority order (lower number = checked first). The first rule that matches wins. See switchboard/lane/policy.py for the schema and switchboard/lane/defaults.py for the built-in policy.

To override routing for a deployment, supply a custom policy YAML to LaneSelector.