Optimize AI model routing portfolio

What it's for

Chooses which model handles which workload against quality, latency, residency, capacity and budget limits at once, instead of defaulting everything to the largest model.

Gives CTOs and investors a finance-aware multi-model routing policy that balances quality and latency with data sovereignty, capacity, common-provider failure and concentration tail risk.

What you give it

Inputs split into evidence read from your connected systems, calibration your team owns, and numerical controls that affect precision but never the result's meaning.

{
  "implementation_budget": 5,
  "maximum_provider_concentration_fraction": 0.6,
  "model_endpoints": [
    {
      "capacity_scenarios": [
        200,
        200
      ],
      "data_retention_compliant": true,
      "evidence_verified": true,
      "id": "model-a",
      "provider_id": "provider-a",
      "serving_region": "eu"
    },
    {
      "capacity_scenarios": [
        200,
        200
      ],
      "data_retention_compliant": true,
      "evidence_verified": true,
      "id": "model-b",
      "provider_id": "provider-b",
      "serving_region": "eu"
    }
  ],
  "routing_options": [
    {
      "available_scenario_ids": [
        "base",
        "provider-a-outage"
      ],
      "dependency_option_ids": [],
      "endpoint_ids": [
        "model-a"
      ],
      "evidence_verified": true,
      "exclusion_option_ids": [],
      "gross_value_scenarios": [
        100,
        100
      ],
      "id": "route-a",

What you get back

This is the actual output of running the example above — computed by the same function the platform calls, not an illustration.

{
  "assumptions": [
    "Every workload selects one prospectively evidenced route whose endpoint shares sum to one. Request volume, value, operating and quality loss, latency, endpoint capacity and provider-failure scenarios share one horizon.",
    "Residency, retention, quality, latency, high-impact provider diversity and endpoint capacity are hard feasibility conditions. Provider concentration uses value-weighted traffic exposure, not endpoint counts or vendor labels.",
    "A provider failure removes every endpoint share on that provider in the same scenario; route unavailability removes the whole route. Gross value and quality loss scale to surviving traffic, while direct workload value at risk is lost only once. Gross opportunity, quality loss, operating cost and direct failure loss must be finance-owned, non-overlapping definitions.",
    "Exactness covers only the submitted finite model and beam output is heuristic. Selection is not a vendor SLA, procurement approval, data-transfer authorization, guaranteed quality/cost or judgment about a provider, team or person."
  ],
  "baseline_current_state": {
    "conditional_value_at_risk": 110,
    "expected_net_value": 30,
    "expected_total_loss": 40,
    "feasible": false,
    "provider_concentration_fraction": 1,
    "routing_option_ids": [
      "route-a"
    ]
  },
  "constraints": {
    "implementation_budget": 5,
    "maximum_cvar_total_loss": null,
    "maximum_endpoint_overload_probability": 0,
    "maximum_expected_operating_cost": null,
    "maximum_expected_total_loss": null,
    "maximum_latency_breach_probability": 0,
    "maximum_provider_concentration_fraction": 0.6,
    "maximum_quality_breach_probability": 0,
    "maximum_route_unavailability_probability": 0,
    "minimum_distinct_providers_for_high_impact": 2,
    "risk_aversion": 0,
    "tail_probability": 0.1
  },
  "decision": "ai_model_routing_portfolio_supported",
  "endpoint_diagnostics": [
    {
      "endpoint_id": "model-a",
      "maximum_selected_load": 50,
      "minimum_capacity": 200,
      "provider_economic_exposure_fraction": 0.5,
      "provider_id": "provider-a"
    },
    {
      "endpoint_id": "model-b",
      "maximum_selected_load": 50,
      "minimum_capacity": 200,

How it works

Constrained optimization — Pick the best feasible option under real limits — budget, headcount, dependencies, capacity — rather than ranking a list and hoping it fits.

1. 1 Freeze workload value and requirements, versioned endpoint capacity and provider identity, coherent scenarios and executable route alternatives with prospective economics and current-state labels.
2. 2 Aggregate traffic from all selected workloads upward into endpoint capacity; remove all shares on failed providers; scale gross value and quality loss to surviving traffic; union breach scenarios and value-weight provider concentration.
3. 3 Enumerate the multiple-choice portfolio exactly when tractable or disclose deterministic beam search, apply hard governance and tail-loss constraints, compare with current state, and return the non-dominated value/CVaR/concentration frontier.

Before you trust it

Every tool in the catalog ships with the conditions under which its answer is meaningful — and the conditions under which it should abstain instead of guessing.

Calibrating it to your org

Calibration workflow

1. 1 Define the management decision, target outcome, aggregate unit, privacy boundary, cadence, and prediction/intervention horizon for this organization.
2. 2 Build a tenant-scoped historical cohort using only information available before each prediction or decision; preserve zero periods, censoring, assignment probabilities, and unresolved outcomes when the method requires them.
3. 3 Estimate statistical parameters on training history, but obtain costs, utilities, risk tolerance, practical-effect thresholds, capacity, and policy constraints from accountable decision owners.
4. 4 Validate on later time windows or held-out aggregate units at the deployment grain, against a simple baseline and the function-specific validation strategy.
5. 5 Deploy only if the returned decision clears evidence, overlap, calibration, robustness, and guardrail checks; warning, unsupported, schema-gap, and fallback decisions are abstentions.
6. 6 Monitor realized outcomes, data drift, coverage, and decision regret; recalibrate at a governed cadence or after a detected regime/definition change, never merely because a stakeholder dislikes the result.

Call it from your AI

You don't wire up 388 tools in your MCP client. The GitRevio MCP server exposes 18 tools, three of which let an agent search the catalog, read a tool's schema, and run it — so the assistant finds this one on its own.

gitrevio_capabilities_search
  { "q": "choose one evidenced aimodel route per" }
  → finds "optimize_ai_model_routing_portfolio"

gitrevio_capability_describe
  { "capability_id": "optimize_ai_model_routing_portfolio" }
  → returns the input schema and agent guidance shown on this page

gitrevio_capability_run
  { "capability_id": "optimize_ai_model_routing_portfolio", "arguments": { ... } }
  → returns the result shown above

Works in Claude Desktop, Claude Code, Cursor, Cline, Continue.dev, Goose and Aider. See the MCP server.

Related tools

See every tool in AI cost, routing & return →