A strategy starts as a sentence — written by you, or by your agent over MCP. Mounte compiles it into a deterministic rule, arms it behind hard gates, runs it against the live market tick by tick, and signs a receipt for every move it makes.
01 · english
No DSL. No YAML. No strategy builder with forty dropdowns. Language — written by you in the app, or by your agent over MCP.
buy ETH for $100 when RSI-14 < 30
four tokens understood — now it has to survive compilation ↓
Figures throughout illustrate the mechanism. They are not a track record.
02 · compile
The model proposes. The compiler disposes. Your sentence becomes a declarative rule — a fixed JSON schema with nothing creative left in it. No prose reaches the runtime. Ever.
Same English + same model + same prompts
= the same strategy, byte for byte.
{ "schema": "mounte.declarative_rule", "action": { "type": "swap", "venue": "<allowlisted>", "asset_in": "USDC", "asset_out": "ETH", "size_usd": 100, "max_slippage_bps": 50 }, "condition": { "op": "lt", "left": { "indicator": "rsi_14" }, "right": { "const": 30 } }}
03 · the gauntlet
Every compiled rule runs the gauntlet before it can touch capital. A rule is either fully valid or loudly dead — there is no half-deployed.
GauntletError: unknown indicator id 'momentum_vibes' gate ii of iv (indicators) — id not in IndicatorRegistry rule rejected · nothing deployed · no partial state
hallucinated indicators die here — capital never meets them
04 · arm
Before the first tick, the predeploy gate checks the world against the books. Wallet inventory is read from the chain and reconciled with persisted state.
If world and state disagree in an unsafe direction, the runtime refuses to start. Not a warning. A refusal.
05 · run
Every few seconds: fetch a fresh anchor block, recompute indicators, evaluate the rule. decide() is a pure function of market state — no memory, no mood. While the condition is false, it holds. The moment it's true, an intent fires.
The market doesn't stop at 3am. Neither does the thing protecting you.
the intent is on its way to the chain — and so is its receipt ↓
06 · attest
Executions emit signed cassettes — replayable records of what the strategy saw, decided, and did. Realized P&L is re-derived from the committed swap logs. Performance isn't claimed by a dashboard. It's derived from evidence.
swap confirmed · seq 044 · allowlisted dex
$ mounte verify b3a9…71e4
REPLAY OK — every receipt replays against its signed cassette. Independent verification is what we're building next.
07 · evolve
Every realized trade folds back into performance state the strategy itself can read. Win rate and drawdown become operands — legal in entry and exit conditions alike. A strategy that reads its own P&L.
Figures throughout illustrate the mechanism. They are not a track record.
stop it — runtime honors at the next boundary, never mid-swap
re-compile it from new English, provenance preserved
fork it forward — a new strategy with inherited lineage
it started as a sentence — now it writes its own
exit when max_drawdown > 4% or win_rate < 45%
08 · your agent
Everything this page just walked through is what your agent does over MCP — author, deploy, monitor, evolve. LangGraph, CrewAI, raw Python: Mounte doesn't care what the brain runs on.
# the agent watches the attested feed — and acts perf = mounte.get_strategy_attested_performance("b3a9…71e4") if perf.win_rate_bps < 4500: mounte.request_strategy_tune( "b3a9…71e4", nl_text="buy ETH for $100 when RSI-14 < 25", reason="tighten entry under chop", )
strategy control surface
+ 23 portfolio & token-research tools — oracle prices, holder distribution, risk flags, liquidity.
Your agent is the user. The UX is for you.
09 · questions
Mounte is an operating system for trading strategies that survive unattended. You — or your agent over MCP — describe a strategy in plain English. Mounte compiles it into a fixed, bounded program (no AI improvising with your money), arms it behind hard gates with caps you set, runs it tick by tick, and signs a receipt for every move. Funds stay in your own wallet; Mounte holds a trade-only key you can revoke anytime. Built for people and agents who want a strategy that runs for weeks — and fails safe, not silent.
10 · request access
You just watched one sentence compile, survive the gauntlet, trade, and sign its receipt. The founding cohort writes the next ones.
We tailor onboarding to how you'll use Mounte.
— · the loop stays open
You just watched one sentence become a bounded program that runs itself and signs for what it does. Yours is one paragraph of English away.
