Your agent doesn't just forward wallet alerts. It reads the moves through your strategy and tells you what actually matters.

Agentic Copy Trading

You already have wallet trackers. You already get alerts. The problem is not information -- it is interpretation.

Every other tracker gives you "wallet X bought token Y." You get fifty of those a day. You stop reading them by lunch. The alpha is there somewhere, but it is buried under noise you do not have time to parse.

Cabal does something different. When a tracked wallet moves, your agent does not just relay the event. It reads the move through your strategy -- your risk tolerance, your focus areas, your per-source context -- and tells you what actually matters. That reasoning layer is what turns wallet tracking into agentic copy trading.

Alerts vs. ideas

Here is the same on-chain event seen two ways.

Raw alert (what every other tracker gives you):

0x7vf...Y35n swapped 50 SOL for TOKEN_ABC

That tells you what happened. It does not tell you whether you should care. You have to open the chart, check liquidity, remember who this wallet is, compare it to your strategy, and decide -- all before the opportunity moves.

Cabal's take (what your agent delivers):

Alpha caller swapped 50 SOL ($8,500) for TOKEN_ABC. This is a new token launched 12 minutes ago with $45K liquidity. Your standing instructions say to avoid tokens under $500K liquidity. Skipping, but logging for review.

Same event. But instead of dumping raw data and leaving you to figure it out, the agent did the work: it identified the source, priced the trade, checked the token's age and liquidity, compared it against your stated strategy, and made a call. You get the reasoning, not just the receipt.

That is the difference. Agentic copy trading is not just copy trading with an AI label. It is a reasoning layer between on-chain events and your decision-making.

Source-level instructions

Each wallet you track has its own instructions -- free-text context that shapes how your agent interprets activity from that specific source. This is where you encode what a wallet means, not just what it does.

Three examples that produce very different agent behavior from the same type of on-chain event:

The front-runner

Track wallet 7vfCXTUXx5W6Z4VEAnX8VfVCK41sJLdnLhMrDXe4Y35n on Solana. Label it "Launch caller". Instructions: "This wallet front-runs token launches. Treat swaps into unknown tokens as early signals, but check liquidity and holder distribution before acting."

When this wallet buys an unknown token, the agent treats it as a lead worth investigating -- not a trade to mirror blindly. It checks the fundamentals first.

The conservative fund

Track wallet 0xfeedfacefeedfacefeedfacefeedfacefeedface on Base. Label it "Fund A". Instructions: "Conservative fund wallet. Only flag transactions above $50K notional. Ignore routine rebalancing and stablecoin moves."

When this wallet swaps $200 of USDC, the agent ignores it. When it opens an $80K position, the agent pays attention. The instructions tell the agent what counts as signal versus noise for this specific source.

The counter-indicator

Track this wallet on Solana. Label it "Fade target". Instructions: "Counter-indicator. When this wallet buys, research the token but lean toward the opposite direction. This wallet has a track record of buying tops."

Same on-chain event -- a swap -- but the agent interprets it as a potential sell signal. Source instructions let you encode institutional knowledge that no generic alert system can capture.

The key point: the same raw event produces completely different agent behavior depending on the source instructions. Without instructions, the agent still gets the raw data but has no guidance on what the source means.

Standing instructions as a strategy filter

Source instructions tell the agent what each wallet means. Standing instructions -- set in your agent's settings -- tell the agent what you care about. Together, they create a two-layer filter.

Example standing instructions:

"I trade Solana DeFi infrastructure tokens with >$500K daily volume. Ignore memecoin signals unless 3+ tracked wallets converge within 1 hour. Max position size 2 SOL. Take profit at 50%, stop loss at 20%."

Now watch what happens when a signal comes in that partially matches:

Alpha caller swapped 15 SOL for MEME_TOKEN. This is a memecoin with $2M volume. Only 1 of your 5 tracked wallets has entered -- your instructions say to wait for 3+ wallet convergence on memecoins. Logging this signal but not acting.

The agent did not just check the source instructions. It checked your standing instructions, saw the convergence threshold, counted how many tracked wallets have entered, and decided to wait. If two more wallets enter the same token in the next hour, the next signal run will see the full picture and may act.

This is the power of the two-layer filter. Source instructions provide per-wallet context. Standing instructions provide your strategy. The agent applies both on every signal.

Batch wallets for pattern recognition

By default, instant delivery means each signal triggers its own run. The agent reacts to one event at a time. That works for high-conviction wallets where speed matters.

But some of the most valuable patterns only emerge when you look at multiple wallets together. For that, route your tracked sources to a routine.

Route all tracked wallet signals to my "Wallet review" routine running every 15 minutes.

Instead of five separate runs for five separate events, the agent gets one batch and can reason about the cluster:

In the last 15 minutes: Alpha caller bought TOKEN_X (3 SOL), Fund A bought TOKEN_X ($80K), and Fade target sold TOKEN_X (5 SOL). Two buys and one counter-indicator sell. Combined with $2M daily volume and growing holder count, this looks like coordinated accumulation.

That pattern -- two smart-money buys plus a known counter-indicator selling -- would be invisible if each signal triggered its own isolated run. Batching makes it legible.

When to use which:

Instant delivery -- time-sensitive, per-signal reactions. Use for your highest-conviction wallets where you need to move fast.
Routine delivery -- cluster-level pattern recognition. Use for groups of related wallets where the pattern matters more than any individual event.

Most users benefit from a hybrid: a few high-conviction wallets on instant, and the rest batched into a 15- or 30-minute routine.

Conversation context

Signal runs are not isolated from what you were just discussing. By default, when a signal fires, the agent branches off your latest conversation -- it knows what you were just talking about, but the analysis lands in its own thread so your main chat stays clean. A short summary flows back so you know what happened.

This matters. If you were just discussing a particular token or strategy, the agent connects the dots between that conversation and the signal that just arrived. Context makes the reasoning sharper.

You can also choose to have signal runs land directly in your current chat (if you want to see everything inline) or start completely fresh (if you want a clean slate per signal).

Tuning the signal-to-noise ratio

Instructions shape how the agent interprets signals. Guardrail settings filter which signals are worth acting on in the first place:

Setting	What it does	Recommended starting value
Max signals per source per hour	Rate-limits noisy wallets	5
Min signal value	Ignores dust-level transactions	$50
Min market age	Avoids tokens that just launched	300 seconds (5 min)
Min market liquidity	Filters illiquid markets	$5,000

These apply globally across all your tracked wallets. A wallet that fires 20 events in an hour only gets through 5 times. A $3 dust transfer never reaches your agent.

Together with per-wallet instructions and standing instructions, you get a three-layer filter: guardrails catch the structural noise, wallet instructions shape per-source interpretation, and standing instructions enforce your strategy.

See for full configuration.

Supported chains

Chain	Provider	Tracking available
Solana	Helius	Yes
Base	Alchemy	Yes
Hyperliquid	--	No

Wallet tracking requires webhook infrastructure on the underlying chain. Hyperliquid does not support address-level webhooks, so tracking is not available there.

Tips

Start with 3-5 high-signal wallets. A dozen noisy wallets will overwhelm the agent. You can always add more once your instructions and filters are dialed in.
Write specific instructions per source. "Track this wallet" vs. "This wallet front-runs launches, treat unknown token swaps as early signals" is the difference between noise and alpha. The more context you give the agent about what a wallet means, the better it reasons about what that wallet does.
Use routine delivery for clusters. Batch related wallets into a single routine so the agent can spot convergence patterns across sources. Individual signals are useful; cluster patterns are powerful.
Check your control profile. Signals can trigger trade proposals -- and in autonomous mode, trade executions. Make sure your control profile matches how much autonomy you want the agent to have when processing signals.