Pensero Agent Traffic Control | Pensero[Skip to content](#main) [Pensero](/) - Platform Measure - [Next movesWhat to improve, why, and what to do next](/platform/next-moves) - [Agentic ROICost per unit of delivered work](/platform/agentic-roi) - [AI DeploymentAdoption, output and quality of AI](/platform/ai-intelligence) - [Delivery IntelligenceOutput and flow by source](/platform/delivery-intelligence) Govern - [Agent Traffic ControlUnderstand agent usage and choose models by delivery](/platform/agent-traffic-control) - [BenchmarkingCompare teams, agents and models](/platform/talent-benchmarking) - [Code QualityDefects and rework by source](/platform/code-quality) - [CapEx & FinanceCapitalization and spend allocation](/platform/capex-analysis) [Pensero MCPNewYour next moves in any assistant](/platform/mcp) - Use cases By role - [CEOs](/use-cases/roles-ceo) - [CTOs](/use-cases/roles-cto) - [CFOs](/use-cases/roles-cfo) - [Product leaders](/use-cases/roles-product-leaders) - [Engineering managers](/use-cases/roles-engineering-managers) - [Investors](/use-cases/roles-investors) Compare Pensero - [DX](/use-cases/competitors-dx) - [Swarmia](/use-cases/competitors-swarmia) - [Haystack](/use-cases/competitors-haystack) - [Span](/use-cases/competitors-span) - [Jellyfish](/use-cases/competitors-jellyfish) - [Waydev](/use-cases/competitors-waydev) - [LinearB](/use-cases/competitors-linearb) - [Faros AI](/use-cases/competitors-farosai) - [Larridin](/use-cases/competitors-larridin) - [Weave](/use-cases/competitors-weave) - [Exceeds AI](/use-cases/competitors-exceeds-ai) - [How it works](/how-it-works) - [Customers](/customers) - [Pricing](/#pricing) - Resources - [About](/about) - [BlogNews, tips, and dev culture](/blog) - [Help CenterGuides for setup, integrations and security](https://help.pensero.ai) [Log in](/auth/login/)[Book a call](/book-demo) [Log in](/auth/login/)[Book a call](/book-demo) Pensero Agent Traffic Control # Rational model usage for coding agents Measure the cost per delivered point for each owner. Pensero keeps premium models where they deliver and routes the others to lower-cost models. [Book a call](/book-demo) Coding agentsOn-prem traffic controlYour model provider platform-teamPass unchangedbilling-svcPremium → lower-cost modelwarehouse-teamPass unchangedPensero proxyOwner policy → modelPremium modelStandard modelFast modelrouted: $/pt above targetrequests →pass unchanged → Coding agents Proxy-compatible clients ↓ requests Pensero proxy Check owner policy ↓ pass, or route premium → lower-cost model Premium modelStandard modelFast model Your model provider Routed: $/pt above target, with enough PR delivery ## What your agent spend delivers Session output Session cost by output type · Last 14 days $246 · 11 sessions PR · 6$171 Doc · 2$12 Research · 1$18 Ops · 1$21 Nothing · 1$24 ![Session cost by output type · Last 14 days for Varrow Freight Group. $246 · 11 sessions.](/assets/agent-types.D8r75Lo2_1IrL7K.webp) ### See where agent spend goes Compare session cost by output type before you change how a team works. Cost and delivery Cost per delivered point $20/pt default target Target · $20/pt $7.20 $27.00 no PR points · flagged platform-teambilling-svcwarehouse-team ![Cost per delivered point for Varrow Freight Group. $20/pt default target.](/assets/agent-owners.DZGULQKi_Z2d1PbJ.webp) ### Keep the model where it earns its cost Compare cost per point per owner to decide where an expensive model is worth keeping. PR scoring PR size × code-level complexity 6 scored PR sessions · 10 points SizeLevel 1Level 2Level 3**S · 1 pt**31—**M · 3 pts**11—**L · 5 pts**——— - Scored sessions · count in each cell ![PR size × code-level complexity for Varrow Freight Group. 6 scored PR sessions · 10 points.](/assets/agent-sizes.CYAcLR0G_Z1D8tAU.webp) ### See the work behind the points PR size sets delivery points; code-level complexity adds context, not more points. Model routing Recent model requests Policy applied to each request **Billing service**$/pt above target Premium → Lower-cost modelRouted **Platform team**Within target Premium → PremiumPass **Warehouse team**Flagged only Premium → PremiumPass **Billing service**$/pt above target Premium → Lower-cost modelRouted ![Recent model requests for Varrow Freight Group. Policy applied to each request.](/assets/agent-routes.k32ryZvc_Z4rEpW.webp) ### See which requests changed model Route premium requests to a lower-cost model; investigate low delivery without blocking models. ## Impact-based routing Compare the cost of token usage with the tasks that sessions delivered. Use USD per delivery point ($/pt) for each owner over a configurable 14-day window. Incoming request**Owner + model**billing-svc · Premium → Owner rating**Check the rating system**billing-svcUnavailable → pass → Owner state**Check cost vs delivery**$27/pt > $20 targetLow delivery → pass, flagged only → Requested model**Premium + fallback state?**Premium modelYes → fallback · otherwise pass Pass unchangedFallback · lower-cost modelrequest Pass unchangedrequest → Fallback · lower-cost model ## Runs in your VPC Your session data stays on your host. Send agent requests to any OpenAI-compatible endpoint or LiteLLM provider, including OpenAI, Gemini, vLLM and Ollama, or Bedrock or Anthropic. Your host or VPC ### Records and control stay with you ProxyLocal LayaScorerDashboardSession filesLocal PR scoring model (optional) One Docker Compose deployment. A proxy key and a password-protected dashboard. → Agent requests + context **Your model provider** → PR scoring + code context **PR scoring model (cloud or local)** [Discuss your on-prem deployment →](/book-demo) ## Questions about agent traffic control Which coding agents work today? Scored sessions today come from Claude Code. The proxy supports any OpenAI-compatible endpoint or LiteLLM provider, including OpenAI, Gemini, vLLM and Ollama, or Bedrock or Anthropic. Routing matches requested model names that contain opus and uses Sonnet as the default fallback. You can configure a fallback from any provider. PR scoring uses Qwen on Bedrock or local Ollama. Other agents can connect through compatible APIs, but their sessions are not scored yet; API compatibility does not mean that every client tool message can be scored. Do delivery points measure business impact? No. Points measure PR-sized output, not business impact, production success or code quality. PR is an output classification, not proof of an opened pull request, merge or deployment. Use the data with a review of the work, not as a complete measure of value. Do not use it to rank people or dismiss non-PR work. How does it count delivery points? PR size determines points: Z 0, XXS 0.15, XS 0.5, S 1, M 3, L 5 and XL 8. Complexity does not change points. Doc, research, ops and nothing earn no points, but their costs count in the owner rating. Ratings compare USD per delivery point ($/pt) per owner (a person, team or service key) over a configurable 14-day window. What does a session record include? The owner, models, backend, tokens, cost, output type, files changed, tool calls and recorded commit SHAs. Records come from supported tool messages, not a full filesystem audit. Repeated edits to existing files can count the same lines more than once. When is a session scored? When the session ends, or on demand from the admin dashboard. The session record supplies the work for local Laya classification. The PR scoring model evaluates cleaned diff, summary, features, magnitude and code-level complexity. Other types get no size and no complexity. Which routing settings can we change? Impact-based routing compares token spend with delivered tasks as USD per delivery point ($/pt) per owner. You can configure the rating window, minimum session count, spend thresholds and fallback model. The defaults require 3 scored sessions in 14 days. At least 1 point and cost above $20 per point routes premium requests to the configured lower-cost model. Below 1 point and cost above $40 flags low delivery only. These settings apply across owners; this is not an arbitrary model-routing rule builder. What happens when a rating is unavailable? Missing, unreadable or expired ratings leave the requested model unchanged. A low-delivery flag also leaves it unchanged. Only a premium-model request from an owner marked for fallback uses the lower-cost model. What does an owner identify? Ratings apply per owner (a person, team or service key). The owner email is a client-supplied label, not an authenticated identity. Use the shared-key proxy only with trusted clients. Does all data stay in our environment? It depends on the endpoints you configure. Agent requests and their conversation and code context go to the model provider or OpenAI-compatible endpoint that you configure. Qwen scoring sends session and code context to Bedrock unless you use local Ollama. Session artifacts, scores, ratings, the dashboard and Laya stay on your host. How do we deploy it? Book a call to discuss an on-prem deployment on your host or VPC. A single Docker Compose deployment provides the proxy, scorer and password-protected admin dashboard. Session data stays in files on your host. Use a protected network with firewall and TLS controls; the deployment has no built-in TLS. ## More on Pensero's platform - [Next moves →What to improve, why, and what to do next](/platform/next-moves) - [Agentic ROI →Cost per unit of delivered work](/platform/agentic-roi) - [AI Deployment →Adoption, output and quality of AI](/platform/ai-intelligence) - [Delivery Intelligence →Output and flow by source](/platform/delivery-intelligence) - [Benchmarking →Compare teams, agents and models](/platform/talent-benchmarking) - [Code Quality →Defects and rework by source](/platform/code-quality) - [CapEx & Finance →Capitalization and spend allocation](/platform/capex-analysis) - [Pensero MCP →Your next moves in any assistant](/platform/mcp) ## Choose the next model with evidence Discuss an on-prem deployment to understand agent usage, delivery and model spend. [Book a call](/book-demo) © 2026 Product shots use illustrative data. Platform - [Next moves](/platform/next-moves) - [Agentic ROI](/platform/agentic-roi) - [AI Deployment](/platform/ai-intelligence) - [Delivery Intelligence](/platform/delivery-intelligence) - [Benchmarking](/platform/talent-benchmarking) - [Code Quality](/platform/code-quality) - [Agent Traffic Control](/platform/agent-traffic-control)New - [CapEx & Finance](/platform/capex-analysis) - [Pensero MCP](/platform/mcp)New - [Integrations](/integrations) Use cases - [CEOs](/use-cases/roles-ceo) - [CTOs](/use-cases/roles-cto) - [CFOs](/use-cases/roles-cfo) - [Product leaders](/use-cases/roles-product-leaders) - [Engineering managers](/use-cases/roles-engineering-managers) - [Investors](/use-cases/roles-investors) - [Compare Pensero](/use-cases/competitors-dx) Company - [About](/about) - [How it works](/how-it-works) - [Consulting](/consulting) - [Careers](/careers) - [Blog](/blog) - [LinkedIn](https://www.linkedin.com/company/penseroai/) - [Support](/support) - [Help Center](https://help.pensero.ai) - [Other resources](/blog/other-resources) - [Glossary](/glossary) Offices - [Barcelona](https://maps.app.goo.gl/6pfxKiKFNKzLPPnB6) - [Madrid](https://maps.app.goo.gl/jEpYBUR4E6AqZs4R6) - [New York](https://maps.app.goo.gl/szk58Jf2MBq14MgAA) - San FranciscoSoon Legal - [Privacy policy](/privacy-policy) - [Cookie policy](/cookie-policy) - [Terms of service](/terms) - [DPA](/dpa) - [Security](https://pensero.trust.site)