Reference strategies

The OSS build ships three strategies as reference implementations. Together they exercise every code path in the runtime so you have a working example for each shape of strategy you might want to write.

Strategy	SwapIntent	OrderIntent	Cancels	LLM veto	Lives at
`noop`	--	--	--	--	`strategies/noop/`
`dca_buy`	✓	--	--	--	`strategies/dca_buy/`
`market_maker_basic`	--	✓	✓	✓	`strategies/market_maker_basic/`
`alpha_dca`	✓	--	--	--	`strategies/alpha_dca/`
`alpha_momentum`	✓	--	--	--	`strategies/alpha_momentum/`
`alpha_yield`	✓	--	--	--	`strategies/alpha_yield/`

`noop`

The smallest strategy that satisfies the SAPI. ~20 lines of Go that return Decision{} every tick. Use it as a copy-paste starting point or as the smoke-test target for verifying your install -- see running noop.

`dca_buy`

Deterministic dollar-cost-averaging into a configured spot asset. No shorting, no LLM. Useful in real life: anyone who wants to DCA into SOL / ETH / WIF from their own wallet has a working strategy out of the box.

Configuration

Pass via --config-json at agent-create time:

{
  "asset": "SOL",
  "chain": "solana",
  "usdc_per_tick": "50",
  "interval_secs": 14400,
  "slippage_bps": 50
}

Key	Default	Notes
`asset`	`"SOL"`	Spot symbol to buy. Resolved against the configured swap router.
`chain`	`"solana"`	Settlement chain. Must have a USDC mapping (Solana, Ethereum, Base, Avalanche, BSC).
`usdc_per_tick`	`"50"`	USDC notional to buy each cycle. Must be positive.
`interval_secs`	`14400` (4h)	Minimum gap between buys. Cooldown after a successful tick.
`slippage_bps`	`50`	Per-swap slippage tolerance. 0..1000.

Behaviour

tick → cooldown active?            ─ yes ──► skip (note: "cooldown: 3h12m remaining")
       │
       no
       │
       ▼
       USDC mapped for chain?      ─ no ──► skip (note: "no USDC mapping for chain X")
       │
       yes
       │
       ▼
       emit SwapIntent: USDC → asset, BasisKey="dca:<ASSET>"
       update lastBuyAt
       Decision{Confidence: 1.0, Notes: "dca buy: 50 USDC → SOL on solana"}

Cooldown is in-memory. A daemon restart resets it; the strategy may buy immediately on the first tick after restart even if it bought recently. v2 will read prior buy times from BasisPositions.

Run it

permafrost agent create \
    --strategy dca_buy \
    --perp hyperliquid \
    --spot solana \
    --alloc 1000 \
    --tick-secs 60 \
    --config-json '{"asset":"SOL","usdc_per_tick":"50","interval_secs":14400}'
permafrost agent start <id>

`market_maker_basic`

Places paired bid/ask limit orders on Hyperliquid around the mid, half-spread = SpreadBps/2 on each side. Refreshes every RefreshTicks ticks. Optional LLM veto consults the inference provider per refresh cycle (JSON-Schema response, graceful degrade on ErrUnsupportedFeature).

Configuration

{
  "symbol":         "WIF",
  "spread_bps":     25,
  "order_size":     "10",
  "refresh_ticks":  1,
  "use_llm_veto":   true,
  "veto_model":     "anthropic/claude-sonnet-4.5"
}

Key	Default	Notes
`symbol`	(required)	Perp symbol (e.g. `"WIF"`).
`spread_bps`	`25`	Half-spread around mid in bps. 1..1000.
`order_size`	(required)	Per-side base-asset quantity. Must be positive.
`refresh_ticks`	`1`	Quote refresh cadence in ticks.
`use_llm_veto`	`false`	If true, ask LLM whether to skip each refresh cycle.
`veto_model`	(agent's)	Falls back to `Services.InferenceModel`.

Behaviour per tick

tick → tickCount++
       │
       ▼
       no market data for symbol?       ─ yes ──► skip (note: "no market data for X")
       │
       no
       │
       ▼
       no mark price?                    ─ yes ──► skip (note: "no mark price; skipping")
       │
       no
       │
       ▼
       refresh-ticks gate                ─ skip ─► skip (note: "skip: tick N, refresh every M")
       │
       │
       ▼
       use_llm_veto and veto = true?     ─ yes ──► skip (note: "vetoed: <reason>")
       │
       no
       │
       ▼
       emit two limit orders: bid at mid*(1-half), ask at mid*(1+half)
       BasisKey = "mm:<SYMBOL>"
       Decision{Confidence: 0.7, ...}

Warmup fails fast if use_llm_veto: true but no inference provider is wired on the agent -- the operator notices at startup, not on the first decision tick.

LLM veto

The strategy POSTs a JSON-Schema-constrained prompt:

System: You are a volatility filter for a basic crypto perpetuals
        market maker. Decide whether the maker should SKIP this
        refresh cycle. Default to NOT vetoing.

User:   Symbol: WIF
        Mark: 1.234
        Funding (annualised): 0.045
        Funding interval: 1h0m0s

        Should the maker SKIP requoting now?

The response must match:

{ "veto": true|false, "reason": "string ≤ 200 chars" }

A provider that returns ErrUnsupportedFeature (e.g. base Ollama model without JSON-Schema mode) is treated as veto: false -- the strategy quotes normally rather than blocking on a feature the model can't enforce.

Run it

permafrost agent create \
    --strategy market_maker_basic \
    --perp hyperliquid \
    --inference openrouter:anthropic/claude-sonnet-4.5 \
    --alloc 500 \
    --tick-secs 30 \
    --config-json '{"symbol":"WIF","order_size":"10","spread_bps":25,"use_llm_veto":true}'
permafrost agent start <id>

`alpha_dca`

Dollar-cost-average into a fixed set of Bittensor subnet alpha tokens. Every N ticks, buys tao_per_buy TAO of each listed subnet's alpha. Pure on-chain — no LLM, no third-party data.

Configuration

{
  "subnets":        [8, 3, 19],
  "tao_per_buy":    1.0,
  "interval_ticks": 10,
  "slippage_bps":   100
}

Key	Default	Notes
`subnets`	`[8, 3, 19]`	Netuids to buy. SN8 = Vanta/Taoshi, SN3 = Templar, SN19 = Inference.
`tao_per_buy`	`1.0`	TAO amount per subnet per buy tick.
`interval_ticks`	`10`	Minimum ticks between buys. With 30s ticks → buy every 5min.
`slippage_bps`	`100`	Per-swap slippage cap (1.00%). AMM slippage on alpha pools can be material.

Run it

permafrost agent create \
    --strategy alpha_dca \
    --spot bittensor \
    --tick-secs 30 \
    --config-json '{"subnets":[8,3,19],"tao_per_buy":1.0,"interval_ticks":10}'
permafrost agent start <id>

Requires:

A funded Bittensor wallet in the keystore (permafrost wallet generate --chain bittensor)
bittensor.allow_submit: true in config.yaml to enable real on-chain trading

`alpha_momentum`

Tracks rolling price changes across all subnets in universe, rotates into the top-K by momentum, exits when momentum flips below exit_threshold. Pure price-action — no external signals.

Configuration

{
  "universe":         [1, 3, 8, 19, 64],
  "window_ticks":     30,
  "top_k":            5,
  "exit_threshold":   -0.02,
  "tao_per_position": 5.0,
  "slippage_bps":     100
}

Key	Default	Notes
`universe`	`[1..64]`	Netuids tracked for momentum.
`window_ticks`	`30`	Rolling window length for momentum (last - first) / first.
`top_k`	`5`	Maximum number of subnets held simultaneously.
`exit_threshold`	`-0.02`	Momentum below which positions are exited (e.g. -2%).
`tao_per_position`	`5.0`	TAO notional allocated to each new position.
`slippage_bps`	`100`	Per-swap slippage cap.

Behaviour per tick

tick → ingest current price for every universe symbol
       compute momentum = (last - first) / first over window_ticks
       sort subnets by momentum desc
       │
       ▼
       held positions with momentum < exit_threshold → emit sell (alpha → TAO)
       │
       ▼
       top-K subnets with momentum > 0 not yet held → emit buy (TAO → alpha)
       │
       ▼
       Decision{Confidence: 0.7, Notes: "holding N subnets, M entries, K exits"}

Position sizing: entries use tao_per_position TAO; exits sell the alpha amount estimated at entry (tao_per_position / entry_price). When the framework surfaces actual venue alpha balances via Services, the strategy will pick those up automatically.

Run it

permafrost agent create \
    --strategy alpha_momentum \
    --spot bittensor \
    --tick-secs 30 \
    --config-json '{"universe":[1,3,8,19,64],"top_k":3,"tao_per_position":2.0}'
permafrost agent start <id>

`alpha_yield`

Ranks subnets by price-stability proxy for emission yield: subnets whose alpha price is stable over the rolling window are ranked higher, the assumption being that emissions accumulate value rather than being eroded by churn. Stakes into the top-K, rebalances every rebalance_ticks ticks.

This is a deliberately conservative stand-in for on-chain Emission storage reads (which require runtime-version-specific storage layout work). The strategy still aims at the same goal — yield-stable subnets — and is the cleanest fork target for a real emission-aware variant.

Configuration

{
  "universe":          [1, 3, 8, 19, 64],
  "top_k":             3,
  "rebalance_ticks":   50,
  "min_yield_delta":   0.05,
  "volatility_window": 30,
  "tao_per_position":  10.0,
  "slippage_bps":      100
}

Key	Default	Notes
`universe`	`[1..64]`	Subnets evaluated.
`top_k`	`3`	Subnets held at any time.
`rebalance_ticks`	`50`	Re-evaluation cadence. With 30s ticks → rebalance every 25min.
`min_yield_delta`	`0.05`	Minimum rank-improvement to trigger a rotation (anti-churn).
`volatility_window`	`30`	Lookback for stability score.
`tao_per_position`	`10.0`	TAO notional per position.
`slippage_bps`	`100`	Per-swap slippage cap.

Run it

permafrost agent create \
    --strategy alpha_yield \
    --spot bittensor \
    --tick-secs 30 \
    --config-json '{"universe":[1,3,8,19,64,52],"top_k":2,"tao_per_position":5.0}'
permafrost agent start <id>

v2 plans for the references

dca_buy -- read prior buy times from BasisPositions so cooldown is restart-resilient.
market_maker_basic -- explicit cancel emission for stale quotes (today the runtime's reduce-only handling drops them naturally as new quotes replace them).
A third reference (grid_trader or similar) demonstrating range-bound passive accumulation.

Next steps

Scaffolding a new strategy -- permafrost strategy-new
Decision contract
Services

noop​

dca_buy​

Configuration​

Behaviour​

Run it​

market_maker_basic​

Configuration​

Behaviour per tick​

LLM veto​

Run it​

alpha_dca​

Configuration​

Run it​

alpha_momentum​

Configuration​

Behaviour per tick​

Run it​

alpha_yield​

Configuration​

Run it​

v2 plans for the references​

Next steps​

`noop`

`dca_buy`

Configuration

Behaviour

Run it

`market_maker_basic`

Configuration

Behaviour per tick

LLM veto

Run it

`alpha_dca`

Configuration

Run it

`alpha_momentum`

Configuration

Behaviour per tick

Run it

`alpha_yield`

Configuration

Run it

v2 plans for the references

Next steps