Turn the report into a testable problem.
Extract the affected service, retry path, symptoms, timing, and available evidence without sending the full workspace history.
Why this route: structured extraction, low ambiguityEach step goes to an appropriate AI model.
Modeus breaks a prompt or task into steps. For each step, the router selects an approved API provider and model based on the work, required quality, modality, policy, speed, and cost.
One task, several routes
A billing incident uses different models for different steps.
Routine extraction can use a focused lower-cost model. Ambiguous diagnosis can use a stronger reasoning model. Tests and tools verify the result.
Trace the lease and callback race.
Use repository context, runtime evidence, and relevant Active Memory to compare causal explanations and identify the smallest defensible root cause.
Why this route: ambiguity, code relationships, consequenceProve the fix against the failure path.
Write the regression test, run the affected suite, and package the evidence with the change instead of asking a model to merely judge its own answer.
Why this route: executable proof is availableRouting signals
The router evaluates each step separately.
It does not assign one model to the entire task just because the task has one category.
- Ambiguity
- How many plausible interpretations remain?Higher ambiguity can justify deeper reasoning or a clarifying question.
- Context
- What information is actually required?Relevant memory and files are selected for the step instead of repeated everywhere.
- Tool need
- Can the result be checked by a system?Search, code execution, browser actions, and structured data tools can be routed independently.
- Consequence
- What is the cost of a weak answer?High-impact decisions can receive stronger reasoning and tighter review boundaries.
- Latency
- Does this step sit on the critical path?Independent work can run in parallel while dependent decisions wait for evidence.
Visible decisions
Users can inspect each routing decision.
The workspace shows which model class and provider handled a step, what context it received, and why the router selected it.
See the model class
Distinguish focused, reasoning, code, multimodal, and verification steps.
See what was sent
Understand which files, memories, and task facts shaped a step.
Set deployment rules
Provider, data, cost, and workflow boundaries remain configuration-dependent.
Bring one high-cost workflow