Choose a policy
Review the models, cost priorities, and quality requirements available in your organization’s Router settings.
Larridin Router · part of Developer Intelligence
Route AI coding work
Route AI coding work
to the right model.
Larridin Router matches AI coding work to suitable models and gives your team visibility into routing decisions, model costs, and quality.
Claude Code · Model routing · Cost and quality
Larridin Router
Example · Last 12 weeks · 64 engineers
31%
cheaper than pinned, $18,420 saved over 12 weeks
87 / 100
code quality on routed work, 86 on pinned
SessionRequested → servedDecisionSaved
Maya ChenClaude CodeOpus 5 → Sonnet 4.6Routed$18.62
Jordan LeeOpenCodeOpus 5 → Sonnet 4.6Routed$3.11
Priya PatelCodexGPT-5.6 CodexPassthrough$0.00
Maya ChenClaude CodeOpus 5Pinned$0.00
Sam OrtizClaude CodeOpus 5Subscriptionn/a
Saved compares each request with the model the developer requested. Requests that cannot be priced are counted, never repriced.
Conference offer$5,000 in Router credits for conference attendees.
Claim your creditsTrusted by AI-forward enterprises
Cost and quality together
Understand what each routing decision changes.
Install, restart your agent, keep working. The router scores each request and serves a lower-cost model when the task allows. Everything else stays on the model the developer asked for.
Per-request scoring · the routing policy evaluates the task and chooses among the models enabled for your organization
Track cache usage · compare cache creation and cache reads when evaluating a policy; session behavior depends on the routing configuration
Know the cost basis · review provider credentials, usage charges, and Router pricing in your organization’s setup
Review fallback behavior · review fallback and pass-through behavior for the clients and models configured in your organization
Example coding tasks
Codebase research
Feature development
Bug fixes
Testing
Refactoring
Larridin Router
Classifier
scores the task, per request
Strategy
Larridin V1 · quality 70%
Budgets
per engineer, per team
Served model
Opus 5Frontier
Sonnet 4.6Efficient
GPT-5.6 CodexFrontier
Fable 5Frontier
Runs on your keys
Anthropic · connected
OpenAI
Fireworks
xAI
Illustrative last 30 days · 37,204 requests routed · 145,237 passed through untouched · $4,812 saved against an all-Opus 5 baseline
Router impact
Follow each routing decision to the PR it shipped.
Cheaper tokens are easy to claim. Larridin ties routed sessions to output, quality, defect rate and cost per task across the SDLC, so you know the discount held up in code review.
Side by side with pinned work · output per engineer, code quality, defects and cost per task, on comparable tasks
Signed savings by strategy · each dollar is credited to the strategy that made the call, and a swap that cost more shows negative
Every session attributed · by developer, agent and credential, with subscription traffic priced at list as estimated API value
Explore Developer Intelligence
Router impact
$18,420
saved, 31% below pinned cost
4.6
PRs per engineer per week
87 / 100
code quality, routed work
2.1%
defect rate, routed work
Where AI coding work ran
Routed 64%Pinned 17%Off-router 19%
Jun 1Jun 22Jul 13Aug 3Aug 17
Routed vs pinned, same tasks
Output per engineer4.6 vs 4.4+4.5% ↗
Code quality87 vs 86within noise →
Defect rate2.1% vs 2.4%−0.3 pts ↘
Cost per task$2.68 vs $4.64−42% ↘
Pinned work is the developer's own model choice on comparable tasks. Off-router traffic ran outside Larridin and is shown for scale. Output, quality and defects come from merged PRs and CI, not from the router's own logs.
Your rules
Decide how hard the router trades cost for quality.
One slider sets the quality target. Everything under it is yours: which models are in play, who can spend what, and what a new session starts with.
Changes take effect on new sessions. A developer who needs a specific model for one job pins it with /modelforce, and that session is reported as pinned rather than counted against the router.
Routing strategies
Larridin V1
ActiveCost-efficient routing. Serves a lower-cost model on sessions the classifier scores as safe to downgrade.
$1,072.66 saved · 30d · Quality target 70%
ManageView logs
Alternative strategy exampleExample policy · $211.40 estimated savingsManage
Custom policy exampleIllustrative configuration · Not enabledManage
Larridin V1 · Quality ↔ cost balance
70% quality
Optimize for costOptimize for quality
At 70%, a session is served a cheaper model only when the classifier clears the quality bar with room to spare. Lower values downgrade more sessions. 100% never downgrades.
Strategies route only within the models enabled on the Models page. Switching strategies takes effect on new sessions only.
Quality vs cost
Set the quality target anywhere from 0 to 100%. Move it whenever the mix of work changes; new sessions pick it up.
Choose the models
Enable the set the router may serve, provider by provider. Nothing outside it is ever served.
Spend visibility
Review spend by engineer and team, then choose a policy appropriate to the work and your budget.
Change the defaults
Set the model tier and caching defaults a new session starts with, per team. Nobody has to remember to opt in.
Strategies
Bring the routing strategy you trust.
One strategy routes at a time, and savings are credited to whichever made the call. Enable another and it takes over on new sessions only.
Larridin V1
The default. Routes against a quality target you set and downgrades only the sessions its classifier clears.
$1,072.66 saved · 31,842 requests routed
Alternative strategy example
An illustrative comparison of task-based decisions between a frontier model and an efficient model.
$211.40 saved · 5,362 requests routed
Custom policy example
An illustrative policy comparison across the models enabled for an organization.
Illustrative policy
Policy cards and savings above are illustrative. Review available strategies in your organization. Keep billed provider charges separate from subscription usage valued at equivalent API rates.
Claude CodeOther clientsSetup guide
Route Claude Code through Larridin
Run one command. Claude Code keeps working as it does now, and every session is metered and attributed to you.
Install the routerSwap in one of your keys from Router Keys, then run this in your terminal.
Copy
curl -fsSL https://app.larridin.com/api/router-connect/install \
| sh -s -- YOUR_ROUTER_KEY
Uninstall removes only what the installer wrote. Settings the developer added themselves are left alone.
curl -fsSL https://app.larridin.com/api/router-connect/uninstall | sh
The documented installer configures Claude Code. Confirm other client setup in your organization’s Router settings.
Install
Connect Claude Code to Larridin Router.
Admins issue a router key per developer. Developers paste it into one command. From the next request, their agent's traffic flows through the router under their name.
1Run the command with your router key.It configures routing and installs the Larridin plugin, which adds /modelforce and, if you have no statusline of your own, one that shows what the router is serving.
2Quit and reopen your coding agent.For the VS Code extension, quit VS Code itself.
3Work as usual.Your organization's provider key serves the request unless you are signed in to a personal subscription. Either way the dashboard attributes spend by developer and model.
Your previous settings are copied to ~/.claude/settings.json.bak the first time you install, and re-installing keeps that snapshot.
Read the Router setup guide for requirements and connection verification.
A closer look
How AI coding model routing works
Routing policies help match a coding task to an appropriate model. Compare cost and quality over the same work and time period before deciding whether a policy helps your team.
Connect a coding agent
The documented installer configures Claude Code. Use your organization’s setup instructions for credentials, routing, and connection verification.
Read the comparison
A comparison against a developer-selected model and a comparison against one premium model answer different questions. Keep the baseline visible.
| Coding agent | Setup | Requirements and scope |
|---|---|---|
| Claude Code | Router installer; see the setup guide below | Larridin account and Router credentials. Review the configuration changes before installing. |
| Other coding agents | Confirm availability in your organization’s Router settings | Client support, installation, and routing coverage can differ. |
Questions about Larridin Router
How does routing choose a model?
A routing policy selects among the models configured for your organization. The task, policy, and enabled models determine the available choices; compare the resulting cost and quality before broadening a rollout.
What happens to prompt caching?
A model change can affect cache reuse. Review cached-input and cache-creation usage separately when comparing policies, and use the session behavior documented for your Router configuration.
What happens when a model is unavailable?
Fallback depends on the policy and providers enabled for your organization. Confirm the fallback and pass-through behavior in your setup before relying on it for a production workflow.
How should I interpret the product examples?
The product views on this page use illustrative data to explain the metrics and workflows. For metric definitions, sample calculations, and assumptions, see our measurement methodology.
Start today.
Connect a coding agent and compare routing decisions, cost, and quality. Try Larridin Router on your own repos, or get pricing for your org.