oddkit and Oddie are MCP servers, not another IDE plugin. They follow you to Cursor, Claude Code, ChatGPT, Lovable — wherever you work. Your governance, your canon, your agent. One bill.
Personal launches first · Pro & Team behind it
oddkit and Oddie are MCP servers, not another IDE plugin. They follow you to Cursor, Claude Code, ChatGPT, Lovable — wherever you work. Your governance, your canon, your agent. One bill.
Personal launches first · Pro & Team behind it
oddkit and Oddie are MCP servers, not another IDE plugin. They follow you to Cursor, Claude Code, ChatGPT, Lovable — wherever you work. Your governance, your canon, your agent. One bill.
Personal launches first · Pro & Team behind it
Built by someone who runs 6 million tokens a month through it. If oddkit doesn't survive your workflow, it doesn't ship to ours.
— Klappy, principal maintainer
Canon is the employee handbook for AI operators and the agents they direct. Outcomes-Driven Development is a software methodology built from the ground up around AI's known weaknesses — hallucination, drift, context loss — by structurally requiring validation before assertion. The methodology proxies trust between you, your team, and the model. You inherit the whole library on signup. It grows continuously.
AI will happily write the code you asked for. It won't notice that it just contradicted your last three decisions, ignored your constraints, and silently shipped a regression. oddkit is the layer between you and the chaos.
$ "add caching to the auth flow" Sure! I've added an in-memory cache to the auth handler. Here's the diff: + const cache = new Map(); + cache.set(userId, token); $ "didn't we decide no in-memory state on this service last week?" You're absolutely right! Let me refactor that to use Redis... $ ".... we don't have Redis." My apologies! Let me use the filesystem instead...
$ "add caching to the auth flow" Checking canon before I touch this... ✓ canon/constraints/no-in-memory-state ✓ canon/decisions/auth-2026-03-stateless ✓ available infra: postgres, KV Per your stateless-auth decision, I'll add a KV-backed cache with TTL. Here's the diff, with the canon citation: + // per klappy://canon/decisions/ + // auth-2026-03-stateless + const cache = new KVCache(env.KV); Encoded as decision artifact. Want me to gate it for review?
Three reversed suggestions in eight lines. Each one violated a different constraint already on the books. The AI doesn't know what canon says unless someone hands it canon.
That's the job.
Full platform access. Bring your own Anthropic key.
Oddie agent runtime + canon. The home Klappy and Seth are designing for.
Flat. Unlimited members. No seat counting.
Pro features pooled across the workspace. Richwood-shaped.
oddkit tokens (canon content served from our MCP server) and Oddie runs (the per-run cost of spinning up the agent). That's the entire bill from us. Conversational LLM tokens are inference — paid directly to Anthropic, BYO key or managed at cost + 15%. We never put a margin on the model.
The water you drink isn't the river you sit on. oddkit charges for the platform; Anthropic charges for the thinking. Two bills. Two lines. No blend.
Drag the sliders. The bill updates in real time.
$10 credit exhausted · $4.00 billed above
Inference is separate. Anthropic bills your LLM tokens directly (BYO key) or we pass them through at cost + 15% (managed). We don't charge for conversational tokens — only oddkit usage and Oddie runs.
Anthropic charges what Anthropic charges. We don't hide it inside the subscription. You pay for oddkit; you pay for the inference. Two lines on the invoice. No surprises.
Anthropic's rate, zero markup from us.
You already have an Anthropic account. Drop your key into oddkit. Every inference dollar flows directly from your card to Anthropic's. We never see it and we don't charge for it.
Available on all paid tiers. Required at oddkit Personal and Team.
Anthropic cost + 15% convenience.
One card on file. One invoice. We handle the key, the rate limits, the billing reconciliation. The 15% covers the operational cost of being your billing intermediary — that's it.
Available on oddkit Pro and Team Pro.
Oddie is the operational voice of the methodology. He's the dry, observant, impossible-to-fluster guide who reads the river so the team can navigate it. He notices what the AI just missed. He reports what canon already decided. He doesn't yell. He doesn't gush. He just sees.
Canon says Tuesday. The screenshot says Thursday. Two-day delta on a published date — probably worth a look before it ships.
This thread has looped back to the same question three times in the last four exchanges. ↻ Eddy forming. The decision that would break the loop is upstream's call.
Token rate dropped from 140/s to 60/s over the last ninety seconds. Not a cliff yet, but the slope is steady. Might want to check what changed at the source.
That's the voice. Unflappable. Dry. Observant. Calibrated against the energy of Nate Bargatze tuned down for someone who actually knows the water. He won't tell you what to do. He'll tell you what he's seeing. The decision is yours.
The first ten people to publicly champion oddkit get on the inside track. Apply now, get first access when Personal ships, and a guaranteed Pro seat the moment Pro launches — free for three years (or until Covenynt Venture hits $10K MRR, whichever comes first). After that, fifty percent off for life. This is not a coupon. It's a contract.
First in line for Personal access the day it ships. You're an ambassador; you're at the front of the queue.
Guaranteed Pro seat the day Pro ships. Free for three years, or until we hit $10K MRR — whichever comes first.
When the ambassador window closes, you keep 50% off Pro for the lifetime of your account.
One published case study within 90 days. Logo on the site. Quote on the landing page. You're a named partner.
oddkit is development tooling, one layer upstream of the mission itself. It doesn't draft translations — it accelerates the engineers who build the software that does. The cross-subsidy thesis is borrowed from Klappy's Sovee days: commercial sales pay full freight, mission work pays a fraction, and the math works because that's how the business is designed. Even at a discount, oddkit pays for itself in mission teams — less rework, more leverage per engineer.
Half off every tier — every feature, every credit, every meter rate — for Bible translation organizations, faith-based missions, and open-source maintainers. The rate is funded by commercial customers paying full price by design, not by squeezing margins.
Mission teams still get real engineering ROI from oddkit — less rework, more leverage per developer — so the rate is discounted, not free. The 50% is what makes the investment sustainable for both sides.
Want free usage too? Mission accounts earn double bonus tokens per referral — 2M tokens per conversion instead of the standard 1M, applied straight to your next bill. The free-month threshold (2 conversions) stays the same for everyone.
Email us a short note about who you are and what you're building. We set it up by hand.
Apply for mission rate →Ten percent of every dollar oddkit collects flows to Bible translation work — from every customer, every tier, on top of the mission-rate discounts. The cross-subsidy is the operating model, not a marketing line.
Your subscription doesn't fund a charity arm. It funds the point of the company.
— Mission rate applies to each tier as it launches, starting with Personal. The discount is locked in for life on the day you're approved — you don't lose it when more tiers come online.
Marketing is expensive. Word-of-mouth from a developer who already loves the tool is not. We'd rather pay you for the introduction than spend it on ads.
Every paid account has a referral URL — personal subscriptions and team workspaces alike. Share it however you'd share anything else you like.
When a referred user converts to any paid tier, you earn 1M bonus tokens applied to your next bill.
Two converted referrals = one free month at your current tier. Twelve = one free year.
Cap is 12 free months per rolling year. Credits expire at the end of each billing cycle. No farming, no gaming. Just real referrals.
— Referrals activate alongside each tier. Personal referrals begin when Personal launches; Team workspace referrals (any member can share the link, credit flows to the workspace bill) when Team launches. Mission-rate accounts earn double bonus tokens per referral — see the Mission section.
Because every other tool charges per seat, every two-person team gets shut out of team plans, and the workspace value you actually use — shared canon, pooled usage, admin — has a fixed cost regardless of headcount. Pay once for the workspace. Add anyone. That's it.
Two things, and only two things. oddkit tokens — the canon content that flows through our MCP server when you (or Oddie) make calls. Oddie runs — a $0.10 base fee per agent run that covers the infrastructure cost of spinning up the sandbox. That's it. The conversational LLM tokens — what your model thinks with — are inference, paid directly to Anthropic. We don't see those dollars and we don't put a margin on them.
Correct. If you bring your own Anthropic key, you pay Anthropic directly and we never touch that money. If you use our managed key for convenience, we charge Anthropic's actual cost plus a 15% operational fee — the invoice itemizes both numbers. Inference is inference. We don't pretend otherwise.
Direct tokens are $2 per million. Oddie is $0.10 per run plus $2 per million for the tokens it processes. The dashboard shows you exactly where you are. You can set a cap if you want hard cutoffs. Nothing happens silently.
Opinionated. oddkit is a software-development methodology built from the ground up for the era where AI is on the team. Hallucinations, context loss, drift — these aren't bugs to apologize for; they're constraints the methodology designs around. Validate before claiming. Encode decisions so future sessions don't relitigate them. Gate transitions between modes so you don't ship while still exploring. If you want a blank slate to configure your own framework, oddkit is the wrong tool. If you want a methodology that takes AI-augmented engineering all the way to production, you're our customer.
The opposite. ODD applies Theory of Constraints to AI-augmented work: identify the single largest source of friction in your workflow right now, fix that one, then find the next one. Discipline is applied surgically — only where it earns its keep by removing real waste for AI operators. Steps that don't pay back get retired. The methodology gets leaner over time, not heavier. It's not "best practices sprinkled everywhere." It's the smallest set of rails that keep the largest leaks plugged.
Same reason we won't ship code we haven't validated. The platform itself — oddkit MCP, canon, the meter — is in production today, but we won't sell access until accounts, payments, and provisioning are wired end-to-end. Personal launches first. Pro adds Oddie's runtime (gated by cost-validated Project Think instrumentation). Team adds workspace billing and shared-canon access controls. We're shipping in the order we can promise honestly. ODD applied to our own launch. The waitlists are real; we'll email you the moment your tier opens.
If your client speaks MCP, oddkit speaks back. We test against Cursor, Claude Code, Claude.ai, ChatGPT, Codex, Lovable, VS Code, and custom agents — and anything else MCP-compatible should work. Let us know if it doesn't.
Canon is the employee handbook for humans and AI agents working together — six-hundred-plus documents that encode how to ship software when the team includes a model. Decisions, observations, constraints, learnings, handoffs, encoded artifacts. It's public at klappy.dev — you read it before you buy. You inherit the whole thing on signup. It grows continuously — every lesson the team learns at six million tokens a month lands as new canon, and your subscription comes with every update. No version pinning, no migration tax. Or fork it. oddkit reads any markdown canon you point it at — yours, ours, somebody else's. The system is the methodology, not the database.
Otters use tools, they read currents the rest of us miss, and they float through chaos with both hands full. Oddie reads your repo like a river — current, eddies, kept-rocks, driftwood — and reports what he sees. The metaphor wasn't assigned. It emerged.