slithy
toves.dev
Focus on Your Domain. We Handle the Plumbing.
Stop writing the same 860 lines of conversation infrastructure. SlithyToves provides the battle-tested foundation so your team can build what actually matters—your domain expertise.
class GreeterPhase implements Phase {
async execute(context: ExecutionContext) {
const result = await streamText({
model: openai('gpt-4o'),
messages: context.state.messages
})
return context.control.stream(result)
}
}The Problem
Every project reinvents the same wheel
Teams waste weeks building conversation infrastructure instead of domain expertise
- 800+ lines per project
- State management, truncation, error handling, provider abstraction—rewritten every time.
- 80% infrastructure time
- Teams spend most time on plumbing, not building what makes their product unique.
- Same patterns, every time
- Conversation loops, token optimization, retries—everyone rediscovers the same solutions.
- Token waste by default
- Sending full conversation history every turn burns tokens and slows responses.
- Death by LLM
- Using LLMs to produce this code is dependent on prompt quality and yields fragile results.
The Solution
You write domain logic. We handle everything else.
Use SlithyToves' battle-tested infrastructure. Focus on implementing your domain expertise.
- ~170 lines: Core
- Conversation loop: load → execute → persist. Battle-tested and production-ready.
- ~690 lines: Plugins
- Storage, observability, token optimization (75-85% reduction), retries, and error handling.
- ~650 lines: Your Domain
- Your phases, your domain expertise, your competitive advantage.
- Production-proven
- Powering AskKev in production agriculture research.
- 2 weeks to launch
- From concept to production—focus on domain logic, not infrastructure.
Case Study
Proven in production
AskKev is a production agricultural research assistant. It is built on SlithyToves' patterns.
Code Distribution: 57% Infrastructure / 43% Domain
Infrastructure
Domain Logic
With SlithyToves, you get the 850 lines of infrastructure for free. Just write your 650 lines of domain logic.
Key Metrics
- → 78.5% token reduction (15,000 → 3,225)
- → <2 weeks to production
Production Stack
- → Cloudflare Workers + Durable Objects
- → D1 (SQLite) for persistence
- → Vercel AI SDK for streaming
- → Multi-phase orchestration
Key Capabilities
- → Intent classification & routing
- → Context-aware research
- → Automatic token optimization
- → Error handling & retries
"SlithyToves didn't just help me write better code; it allowed me to act autonomously. We shipped a production-ready foundation in the time it usually takes to debate a folder structure."
Velocity Report 2025-11-23
Start (Empty Folder) 09:26 AM
Finish (Production Ready) 09:59 AM
Total Time 33 Minutes
— Excerpt from Case Study: From Zero to OpenChat