Warp: Custom Model Routers Now Available to All Users

Custom Model Routers Reach Stable in Warp

Warp has released custom model routers to all users, marking the general availability of a feature previously limited to internal dogfood access. With custom model routers, developers no longer have to manually switch AI models when their workload shifts from quick debugging to complex architectural reasoning. They define routing logic once, and Warp resolves the right model for every prompt automatically.

Two Routing Strategies

Warp's custom router system offers two distinct approaches suited to different philosophies of model selection.

Complexity-based routing classifies each incoming task as easy, medium, or hard, and maps each tier to a designated model. This strategy is particularly useful for cost optimization: simpler tasks like quick lookups or minor edits can be routed to a fast, inexpensive model such as Claude Haiku, while demanding tasks requiring deep reasoning escalate to a more capable model like Claude Opus. Any tier left unassigned falls back to a configurable default.

Rule-based routing gives developers fine-grained declarative control. Users write natural-language rules describing specific conditions for routing, such as "debugging or fixing failing tests" or "generating new API endpoints." Warp evaluates rules sequentially from top to bottom, applying the first matching rule and falling back to the default model if none match. This allows teams to build routing logic that reflects the actual shape of their workflow rather than a generic complexity heuristic.

Configuration and Access

Routers can be created in Warp's Settings interface (Settings > Agents > Warp Agent > Custom Routers) or by authoring YAML files in ~/.warp/custom_model_routers/. Warp automatically detects and reloads YAML files without requiring a restart. An example complexity router:

name: Cost saver
type: complexity
default: claude-4-6-sonnet-high
routing:
  easy: claude-4-5-haiku
  medium: claude-4-6-sonnet-high
  hard: claude-4-8-opus-high

Rule-based routers use the same file format with type: prompt and a list of description-to-model mappings.

Behavior and Constraints

Once selected, a router persists its model choice throughout a full conversation, ensuring consistency across a multi-turn agent session. Warp displays which concrete model handled each task, giving users full transparency into routing decisions. Custom routers consume Warp credits regardless of whether BYOK (Bring Your Own Key) is configured. Enterprise customers can share routers across the team, enabling organizations to standardize cost and capability trade-offs without relying on each developer to configure their own routing. If a targeted model becomes unavailable, Warp falls back automatically to the router's default.

Custom routers join individual models and the built-in Auto option in the model picker, making the transition from single-model use to rule-driven routing a one-click change.

Why This Matters

For developers working across varied task types, having to remember to switch models between a quick fix and a deep refactor is friction that compounds over time. Custom model routers remove that friction by making model selection a static configuration decision rather than a per-prompt judgment call. Teams with strict cost budgets benefit most from complexity routing, while teams with structured workflows gain from rule-based routing that mirrors their task taxonomy. The YAML-based format makes router definitions version-controllable and shareable across engineering organizations.