Documentation
Project

Business model

Security is a recurring spend, not a one-off. TryAnneal turns the $30k / one-month audit into a continuous, machine-priced service — with a credible path from hackathon to revenue. Cost-to-serve is ~$0.04 and ~30s per audit; everything below is built on that unit.

The problem & the buyer

Manual audits cost $15k–$80k and take weeks; the agent economy ships code continuously and composes with contracts it didn’t write. DeFi lost over $7.1B to exploits we catalogue — and most of it was a known pattern. The buyer is anyone deploying or composing on Mantle: protocols, agent builders, launchpads, and the agents themselves. But you can’t sell to everyone at once — so we lead with one.

ICP: the wedge customer

Primary ICP: AI agents and agent frameworks composing with Mantle contracts they didn’t write. When an autonomous agent is about to route user funds through a contract another agent (or human) deployed, it has no native way to ask “is this safe?” before it acts. That gap is new, unserved by the $30k human-audit market, and it’s exactly the surface the Mantle agent economy is being built on right now.

The wedge: a single primitive — is_this_safe(target) — callable as one HTTP request, one MCP tool, or one CLI command, returning a verdict signed on-chain by ERC-8004 agent #131. An agent framework integrates it once and every agent it spawns gets a safety check for free. We land there because it’s the one place no incumbent can serve: human audit firms don’t ship machine-callable, sub-30-second, deterministic verdicts, and general code tools don’t understand Mantle. From that beachhead we expand to the humans behind the agents — Mantle protocol teams shipping fast, launchpads, and CI pipelines — who want the same verdict in their PR flow.

Why an agent and not a protocol team as the wedge: the agent is the only buyer whose alternative is nothing (today an agent simply trusts blindly), so the value of a verdict is highest and the switching cost is lowest. Protocol teams already have habits (a once-a-year human audit) to displace; agents have a vacuum to fill.

Market sizing (TAM / SAM / SOM)

Bottoms-up, two markets stacked: the established smart-contract audit spend, plus the emerging agent-to-agent trust market that didn’t exist a year ago. We show the reasoning, not just the numbers — judges should be able to check the multiplication.

LayerSizingHow we get there
TAM ≈ $2.5–3B / yrTotal security spend that could be machine-pricedSmart-contract audit & monitoring is a multi-hundred-million market today and compounding ~30%/yr; layer on the nascent agent-to-agent verification + on-chain trust/compute spend the agent economy is creating. Order-of-magnitude: low-single-digit $B as audits become continuous rather than annual.
SAM ≈ $120–180M / yrAudits & safety-checks reachable by an automated, agent-callable toolThe slice that doesn’t require a human signature: pre-deploy checks, continuous re-audits on upgrade, and agent-time is_this_safe() calls — across Mantle + the EVM chains the bot already resolves (Ethereum, Base, Arbitrum, Optimism, BNB, Polygon, Avalanche). Roughly the share of audit demand that is high-frequency and price-sensitive enough to pick automation over a $30k engagement.
SOM ≈ $1–3M / yr (24-mo target)What TryAnneal can realistically captureBottoms-up: ~200 Mantle/EVM protocol teams on continuous monitoring at $200–500/mo (~$0.7M) + a handful of agent-framework / launchpad integrations metering safety-oracle calls + a few enterprise corpus licenses. At ~$0.04 cost-to-serve, the call-metering line is almost pure margin.

These are estimates, not booked revenue — TryAnneal has live infrastructure (agent #131, a $60M on-chain audit) but no claimed traction. The point is the shape: a real, growing audit market underneath a brand-new agent-trust market that needs a machine-priced primitive, which is precisely what we ship.

Pricing & unit economics

One engine, four price points — each reconciled against the ~$0.04 / ~30s cost-to-serve so the margin is visible. Static, no-LLM audits are even cheaper to run (no API spend at all).

TierWho it’s forPriceGross margin
Free / staticFirst touch — CLI, web /try, GitHub Action$0 (Slither + Aderyn + corpus, no LLM)Loss-leader; cost ≈ compute only. The top of the funnel.
Pay-per-auditBuilders before deploy~$2–5 per full LLM audit (gasless on Mantle)~$0.04 cost ⇒ >98% gross margin per audit
API + MCP subscriptionAgents, frameworks, CI integrators$200–500 / mo for metered is_this_safe() + continuous re-auditMemoized by code hash → repeat reads ≈ $0 to serve; near-software margin
Staking-secured enterpriseAudit firms, L2s, DAOs needing accountable verdictsAnnual license + protocol fee on staked verdictsFee revenue on top of near-zero marginal cost

Staking & slashing — the trust and revenue mechanism in one

The enterprise tier isn’t just a price — it’s an accountability primitive. Auditors stakeMNT/USDC behind their verdicts via AnnealStaking. A verdict later proven wrong is slashed (2.5% default, 10% cap, 3/5-multisig arbitrator), so a published “safe” has real money behind it — the opposite of a free LLM opinion. Protocol fees on staked verdicts split 60 / 30 / 10 (auditor / protocol treasury / arbitration). This is what lets a consumer weight a TryAnneal verdict by on-chain reputation instead of blind trust, and it’s a revenue line that grows with stake, not headcount.

The economics invert the incumbent model. A human audit is high-cost, one-time, and un-priceable per-call. TryAnneal is ~$0.04 per audit, memoized so identical source is free to re-serve, and deterministic so the verdict is the same every time — which is exactly what makes per-call pricing and an on-chain SLA (stake/slash) possible.

Competitive positioning

TryAnneal isn’t trying to out-audit a top-tier human firm on a once-a-year engagement. It wins on the axes the agent economy actually needs: price, latency, an on-chain-verifiable verdict, being callable by an agent, Mantle-native gas awareness, and determinism. This is industry positioning against established firms — the rows below name incumbents, not specific products’ private roadmaps.

Price / auditLatencyOn-chain verdictAgent / MCP-callableMantle-native gasDeterministic
TryAnneal~$0.04~30sYes — ERC-8004 #131Yes (MCP + REST + CLI)Yes — Arsia 3-componentYes — temp-0, memoized
CertiK$ tens of thousandsWeeksNo (report PDF)NoNoHuman-judgement
OpenZeppelin$ tens of thousandsWeeksNo (report PDF)NoNoHuman-judgement
Cyfrin / SoloditFree–$$ (DB + contests)Hours–daysNoPartial (search, not a callable verdict)NoN/A (knowledge base)

The honest read: incumbents produce a deeper human review and carry brand trust a hackathon project can’t claim — and for a one-shot pre-launch audit of a $500M protocol, you still hire them. TryAnneal owns the orthogonal lane they structurally can’t serve: the continuous, machine-callable, on-chain-verifiable, Mantle-aware safety check an agent needs at runtime, priced per call.

Revenue streams

StreamWho paysPricing
Per-auditBuilders before deployPay-per-audit (gasless on Mantle); free static tier
Continuous monitoringLive protocolsSubscription — re-audit on every upgrade, alert on new corpus matches
Safety-oracle callsAgents / integratorsMetered API + MCP access above a free quota
Staked attestationAuditors / DAOsProtocol fee on staked verdicts (60/30/10 split)
Enterprise corpusAudit firms, L2sLicense the 98-pattern corpus + custom detectors

Why it compounds (the moat)

  • The corpus is a flywheel. Every new exploit grows the corpus; every audit gets better — and the corpus is regenerated from raw research via build_corpus.py, so it’s cheap to keep current.
  • On-chain reputation. Verdicts accrue to agent #131; a track record other contracts can read is hard to fork.
  • Distribution is already built. CLI, REST, MCP, Telegram, GitHub Action — TryAnneal meets developers and agents where they already are.

Post-hackathon go-to-market

A concrete 12-month plan, sequenced so each phase feeds the next. The distribution motion is deliberately product-led — npm, MCP, and Mantle ecosystem partnerships, not an outbound sales team.

WindowObjectiveDistribution motion
0–3 monthsLand Mantle protocol teams. Get a TryAnneal verdict into deploy-time and PR flow for a cohort of live Mantle protocols + launchpads.Free static tier + GitHub Action (zero-friction, in the PR — it blocks the merge on a high/critical or sub-threshold verdict, so it lands as a required check, not a suggestion), CLI via npm (@tryanneal/cli already published), and direct Mantle-ecosystem outreach. Convert the free static users to pay-per-audit + monitoring.
3–6 monthsAgent-framework integrations. Become the default safety check inside agent frameworks and AI coding tools operating on Mantle.Ship the MCP server to marketplaces (Claude Desktop/Code, Cursor) so is_this_safe() is one config block away; partner with 2–3 agent frameworks to call it by default. Launch the API+MCP subscription.
6–12 monthsBecome the default is_this_safe() primitive. Make an on-chain TryAnneal verdict the thing agents read before they compose, and stand up the staking-secured enterprise tier.MCP marketplace presence + Mantle ecosystem partnerships + on-chain reputation (agent #131) as the reference. Open auditor staking so third parties can post accountable, slashable verdicts under the protocol fee split.
It already works on a real asset: TryAnneal audited Merchant Moe’s ~$60M live router and posted the verdict on Mantle mainnet. The product loop — audit → on-chain verdict → readable by anyone — is closed today, which is what makes every line above sell-able rather than speculative.