3
Pipeline types built-in
<80ms
Routing decision
$20/mo
Base — all models
5.5%
Fee on routed cost
Before any model is called, ModeSwitch evaluates your task across five dimensions. The result: either the single best model, or a coordinated pipeline.
Your task
"Refactor our payment service to support multi-currency"
Decision Engine
Single model
Best model for this task type, dispatched immediately.
Pipeline
Multiple models in sequence, each doing what it's best at.
When pipeline = no
Direct dispatch to the optimal model
Simple bug fix
Fast, precise code edits
Logical reasoning
Best structured thinking
Real-time info
Live web access
Large repo read
2M token context window
// You write this
modeswitch.run("Fix the N+1 query in users.ts")
// We handle this
→ Claude 4.5 Sonnet · single model · 1.2s
When a task requires more than one model can reliably deliver, ModeSwitch composes a pipeline automatically. Each step receives the full output of the previous one—not a summary, not a truncated context.
Defines component boundaries, data contracts, and error handling strategy before a single line of code is written.
Produces production-ready code strictly following the architecture. No improvised decisions. No scope creep.
Reviews for security vulnerabilities, edge cases, and logical errors. Flags issues before you ever run it.
🏗️ Architecture → Generation → Audit
For complex features and critical systems. Each model picks up exactly where the last left off. The auditor sees the full output, not a summary.
Use when
ModeSwitch doesn't wait for the decision engine before generating. A fast draft starts immediately. The engine evaluates in parallel.
If the fast model is sufficient, you get the response instantly. If not, ModeSwitch escalates to a stronger model or triggers a pipeline—without you doing anything.
<80ms
Routing decision
0
Wasted round trips
Auto
Escalation logic
execution_timeline.ts
Draft starts immediately
Fast model generating in background
Decision engine evaluating…
complexity · type · risk · cost
✓ Sufficient
Draft returned instantly. Done.
↑ Escalate
Stronger model or pipeline kicks in.
The fast path costs less. The strong path costs more.
You always get the right one.
Connect a repository and ModeSwitch operates directly on your real code. Describe what you want done. It reads the repo, plans the work, executes it, and optionally opens a PR.
Repo understanding
ModeSwitch reads your entire codebase as context—not just the file you paste. It understands architecture, dependencies, and conventions.
Refactoring assistance
Describe the change you want. ModeSwitch scopes the impact, plans the edit sequence, and applies changes across all affected files.
Bug detection
Not just linting. Multi-model analysis to catch bugs that emerge from interactions between services, race conditions, and logic errors.
Architecture analysis
Ask questions about your codebase. Get answers grounded in the actual code, not generic advice.
Automated changes
ModeSwitch can open pull requests with the changes applied. Review-ready output, not a suggestion to copy-paste.
// Connect your repo — then just describe the task
const result = await modeswitch.run({
task: "Find all places where we're not handling
the case where user.plan is undefined,
and fix them with a safe fallback",
repo: "github.com/acme/api",
output: "pull_request",
});
// ModeSwitch reads the full repo (Gemini 3 Pro)
// Finds all 14 occurrences across 8 files
// Writes the fix (Claude 4.5 Sonnet)
// Audits for regressions (GPT-5.4)
// Opens a PR — ready to review
PR #247 opened — ready to review
14 files changed · 3-model pipeline · audited & validated
Stop guessing which AI to open. ModeSwitch classifies your intent in milliseconds and dispatches to the specialist—at model cost + 5.5% on what the routed model costs.
GPT-5 5.2 / 5.4
OpenAI
Mission-critical logic & factual precision
When an algorithm must work 100% on the first try. GPT-5's Thinking Mode audits complex refactors, multi-step plans, and high-stakes logic with near-human reasoning and minimal hallucinations.
Claude 4.5 Opus / Sonnet
Anthropic
Clean code & software architecture
The most "engineering-minded" model. Doesn't just write code—writes code that feels authored by a Senior Developer, with strict attention to security, modularity, and clean architecture.
Gemini 3 Pro / 3.1
Whole-repo analysis & multimodal tasks
With a 2M+ token context window, it's the only model that reads your entire codebase at once—catching bugs that emerge from interactions across ten files in separate directories.
GPT-5.3 Codex Codex
OpenAI
IDE agents & fast code generation
Massively optimized for rapid, dynamic code generation inside your IDE. Prioritizes response speed and Tool Use—runs terminal commands autonomously so your flow never breaks.
Grok 4.20 4.20
xAI
Real-time data & financial intelligence
Need to know what happened five minutes ago? Grok's architecture is tuned for live data retrieval from the web and social streams—faster than any competitor for real-time market signals.
ModeSwitch routes automatically — or you can pin a model for any conversation.
Every feature is designed around one principle: reduce the gap between describing a task and getting production-ready output.
OpenAI-compatible endpoint. Change one line of code. ModeSwitch handles model selection, fallbacks, and retries automatically.
For tasks that exceed what a single model can reliably deliver. Each step receives full context from the previous—not a summary.
A fast draft starts immediately. The router evaluates in parallel. You never wait for the decision engine to finish before generation begins.
Connect a repo. Describe a task. ModeSwitch reads the full codebase, applies changes across files, and opens a PR when done.
$20/mo base + model cost + 5.5%. Routing picks cheaper models — so token spend drops first, then we take 5.5% of that smaller number. Fee funds itself.
One conversation. All models. ModeSwitch maintains shared memory across model switches so mid-thread context is never lost.
Routing sends each task to the cheapest model that can handle it. Your average token cost drops before we even charge a fee. The $20/mo + 5.5% funds itself from those savings.
01
Task arrives
Router evaluates complexity, type, context window, and cost ceiling.
02
Cheapest sufficient model wins
Simple summary? That's Haiku, not Opus. You never pay Opus prices for Haiku work.
03
Fee is on the cheap model's cost
We charge 5.5% of what Haiku costs — not what Opus would have cost. Fee scales with savings.
04
$20/mo base, all models
Flat base replaces $270/mo in individual subscriptions. The fee funds itself from token savings.
Concrete example — "Summarize this doc"
Without ModeSwitch
Claude Opus
$0.05
per call
routing decides
→
Haiku is enough
With ModeSwitch
Claude Haiku
$0.002
+ fee: 5.5% × $0.002 = $0.00011
We captured 5.5% of $0.002 — not $0.05. You kept $0.048 per call.
Does the fee pay for itself? — 1,000 tasks/month scenario
Without routing
With ModeSwitch routing
$289 saved per month
Routing cuts token spend from $50 → $9.80 — savings alone cover the $20.54 fee. The $270 subscription cancellation is just upside.
$20/mo base — one plan, every model, self-funding fee
| Model / Feature | Individual subscription | ModeSwitch |
|---|---|---|
| GPT-5 (OpenAI) | $30/mo | Pay per use |
| Claude 4.5 Opus/Sonnet | $20/mo | Pay per use |
| Gemini 3 Pro (Google) | $20/mo | Pay per use |
| GPT-5.3 Codex | $20/mo | Pay per use |
| Grok 4.20 (xAI) | $30/mo | Pay per use |
| Smart model routing | ❌ | ✅ |
| Unified conversation context | ❌ | ✅ |
| Cost analytics dashboard | ❌ | ✅ |
| OpenAI-compatible API | ❌ | ✅ |
| Total monthly cost | ~$270+/mo subscriptions for all 5 models | $20 + model cost + 5.5% fee funds itself from token savings |
* Scenario uses 1,000 tasks/month with realistic routing distribution. Token savings vary by workload mix. Subscription prices based on individual Pro/Plus tiers as of 2025.
Limited early-access spots
Describe the task. ModeSwitch routes to the cheapest sufficient model, runs the pipeline, and delivers production-ready output. Early members lock in 4% fee instead of 5.5%.
Routes across
No credit card required · Join 2,400+ developers