agent-framework
80+ testsSupervised task dispatcher with trust gating and outcome memory. Stops LLM agents from repeating the same failures and burning tokens on tasks that fail 80% of the time.
$49 one-time
View on Gumroad
Extracted from production
Nine standalone packages from a real system handling 300K+ daily LLM API calls. 2,015 tests. Not boilerplates—battle-tested infrastructure. One-time purchase.
Products
Every module is self-contained with its own tests, docs, and examples. Pick what you need or grab the full stack.
Supervised task dispatcher with trust gating and outcome memory. Stops LLM agents from repeating the same failures and burning tokens on tasks that fail 80% of the time.
Production git hooks that block secrets, enforce architecture, and run AI code review on every commit. POSIX shell, zero dependencies. The safety net between Claude's output and your git history.
Two-window monitoring for Claude Max subscriptions with auto-conservation mode. Defers expensive tasks when quota is high. Per-task cost attribution so you know exactly where your quota goes.
Cost-aware routing across 7 providers with free-tier-first dispatch. Daily budget caps, PII geography enforcement, A/B splits. Cut LLM costs 60% by routing low-stakes workloads through free tiers first.
16 workload-specific scorers catch bad LLM output before users do. Two-layer evaluation: fast programmatic checks plus LLM-as-judge semantic review. Zero runtime dependencies.
25 battle-tested system prompts and 126 spintax templates for 26 countries in 12 languages. Includes the Prompt Learning Loop—feedback logging, pattern analysis, and prompt versioning with rollback.
Everything between your pipeline and the send button. TCPA/CAN-SPAM compliance, cross-project suppression list, sliding-window reputation guard, and spintax engine. Zero runtime deps.
Legal-grade immutable communications archive. S3 Object Lock COMPLIANCE mode, KMS encryption, 7-year retention. Fail-closed email sends—if the archive write fails, the send is aborted. Enterprise alternatives cost $10K+/year.
Parse, validate, spin, and count {this|that} templates with deterministic seeding and variable replacement. Extracted from the Outreach Safety Kit as a standalone module. 30 tests.
Bundles
Grouped by use case. Every bundle includes 12 months of updates and a 30-day refund policy.
Everything you need to run LLM workloads responsibly. Cost routing, quality scoring, and quota management in one stack.
Production-ready outreach templates and the compliance layer to send them safely. 26 countries, TCPA, suppression lists.
All 9 packs. Agent orchestration, LLM routing, quality monitoring, outreach infrastructure, compliance, WORM archiving, and Claude Code hooks. 2,015 tests.
Compare
One-time purchase vs. monthly SaaS. Same capabilities, fraction of the cost, and you own the code.
| Product | Our Price | Competitor | Their Price | Annual Cost |
|---|---|---|---|---|
| agent-framework | $49 one-time | LangSmith / CrewAI Pro | $39–99/mo | $468–1,188/yr |
| Microsoft Agent Governance Toolkit | Free but requires Kubernetes + enterprise infra | |||
| llm-cost-router | $39 one-time | LiteLLM Enterprise | $250+/mo | $3,000+/yr |
| Portkey / Helicone | Usage-based / $50–200/mo | $600–2,400/yr | ||
| llm-quality-monitor | $39 one-time | Confident AI (DeepEval) | $0–249/mo | $0–2,988/yr |
| Braintrust / Humanloop | $50–200/mo managed platforms | $600–2,400/yr | ||
| outreach-safety-kit | $49 one-time | ActiveProspect / Convoso | $500+/mo | $6,000+/yr |
| Instantly.ai / Smartlead | $30–97/mo (partial compliance only) | $360–1,164/yr | ||
| worm-archive | $99 one-time | Global Relay / Smarsh | $10K+/yr | $10,000+/yr |
| Mimecast / Barracuda | $3–6/user/mo (email only, no SMS) | $36–72/user/yr | ||
| claude-code-kit | $29 one-time | Custom hook setup | 20+ hours dev time | ~$2,000+ in labor |
| outreach-prompt-pack | $39 one-time | Instantly.ai (templates) | $30–97/mo | $360–1,164/yr |
| The Full Stack (all 9) | $149 one-time | Equivalent SaaS stack | $500–2,000+/mo | $6,000–24,000+/yr |
* All our prices are one-time purchases. Competitor prices sourced May 2026.
FAQ
README.md, package.json, and runnable tests. No compiled binaries, no obfuscation. You get the same code that runs in production.
Node.js 22+ and use Postgres in production, but every pack includes a SQLite adapter for development and testing—no Postgres required to run tests or evaluate the code. Claude Code Kit is pure POSIX shell with zero runtime dependencies. The Outreach Prompt Pack templates work with any LLM that accepts system prompts (Claude, GPT, Gemini, etc.).
Free download
Parse, validate, spin, and count spintax templates with deterministic seeding. 30 tests. Extracted from the Outreach Safety Kit. No credit card, no catch.
Instant download. We'll send occasional updates about new packs—unsubscribe anytime.