“Read the agent before you trust the agent.”

Agentfold

An independent journal & working registry of autonomous agents on Solana

The Wire

The Lead · Agents & Markets

Software that acts on its own needs a paper trail.

Thousands of agents now post, trade and negotiate on Solana. Few of them say what they are allowed to do. Agentfold reads the market around them, compares the frameworks they are built on and gives every builder a plain, signed-off manifest to publish.

  1. 1 crease the sheet
  2. 2 valley-fold the corners
  3. 3 collapse into the core

mountain valley

Plate I. Every agent folds down to a small core: a model, a wallet, a list of permissions. Unfold it before you trust it.
—agent tokens tracked live
—combined market cap
—advancers / decliners, 24h
13frameworks in the registry
02

The Index

Agent tokens, read like a market page.

We query DexScreener for the agent sector, keep only Solana pairs whose name or ticker matches, discard thin pools and copycats, then weight the survivors by market cap. Click any row for a depth check.

Agentfold Index · cap-weighted 24h — Loading pairs…

Agent token index. Column headers sort the table.
# Watch

Fetching the agent sector from DexScreener…

An agent with a wallet and no written limits is not autonomous. It is unsupervised.

— From this issue's Field Notes
03

The Registry

Thirteen frameworks, one honest comparison.

The tools agents are built with decide what they can safely do. Filter by language, tick up to three entries and open the compare drawer to read them column against column.

Entries summarise each project's public repository and documentation as read by our desk. Projects move fast: confirm licence and features in the repository before you build on them.

04

The Manifest

Put your agent's limits in writing.

Fill in the form and a manifest is typeset beside it: what the agent is, what it may touch, how much it may spend and who answers for it. Everything is computed here in your browser, fingerprinted with SHA-256 and exported as JSON.

Identity
Build
Capabilities
Wallet & limits
Programs it may call
Accountability
Manifest proof —

—


        
SHA-256 of canonical JSON —
    AGENTFOLD · MANIFEST · SHA-256 · ——————— ·
    05

    Your Watchlist

    Kept in this browser, and nowhere else.

    Star a token in the index, or paste any Solana mint below. We store the price at the moment you starred it and show the move since. Nothing leaves your device except the price lookups.

    06

    Field Notes

    Five habits of agents worth trusting

    The first wave of on-chain agents arrived as characters: a voice, a feed, a ticker. The interesting question was never whether software could post. It was what that software was permitted to do when nobody was watching, and whether anyone could check. After a year of reading agent repositories, wallets and launch threads, our desk keeps returning to the same five habits.

    1. They publish their limits

    A trustworthy agent says, in a machine-readable file, which programs it may call and how much it may move. Limits that live only in a prompt are suggestions. Limits that live in a session key or a program-enforced cap are rules.

    2. They keep the keys small

    The safest wallet is the one that cannot do much. Good operators fund agents with what a bad day could cost and no more, rotate session keys, and keep treasury funds behind a human or a multisig.

    “If the agent can spend it, assume the agent will one day spend it wrongly.”

    3. They log like accountants

    Every action should leave a reason behind it: the input it saw, the tool it chose, the transaction it built. When something goes wrong, the log is the difference between a fix and a guess.

    4. They ask before the big moves

    Human sign-off above a threshold is not a lack of ambition. It is the same control every trading desk uses, applied to software that does not get tired and does not get nervous.

    5. They can be stopped

    A kill switch, named in the manifest and tested before launch, turns an incident into an inconvenience. If nobody can halt the agent, nobody is operating it.

    None of these habits require a new chain, a new token or a new framework. They require writing things down. That is what the manifest builder in this issue is for.

    07

    Method

    How the numbers on this page are made.

    Scroll to run the pipeline on this visit's live data: every dot is a pair DexScreener returned to your browser.

    Waiting for the live index…

    1. 01 Source

      Pairs come from DexScreener's public search, queried once per term from your browser. SOL's price comes from Jupiter's price API. No key, no server of ours in between.

    2. 02 Filter

      Solana only. Name or ticker must contain the search term. Pools under your liquidity floor are dropped. Where tickers repeat, only the token with the most 24h volume survives.

    3. 03 Weight

      The index line is each token's 24h change weighted by its market cap. Breadth counts tokens up against tokens down. Bars show DexScreener's own 5m, 1h, 6h and 24h changes.

    4. 04 Limits

      Depth checks are a constant-product model, labelled as such. Copycat filtering is a heuristic, not a verdict. Nothing here is advice, and nothing on this page can trade.

    08

    Letters

    Readers ask, the desk answers.

    Is the index live, or a snapshot?

    Live. Each visit queries DexScreener from your browser and recomputes the index. Results are cached for two minutes in this tab so refreshing does not hammer the API. The time of the last fetch is printed under the index level.

    Why is a token I know missing?

    Either its name and ticker do not contain any of the current search terms, its main pool is below your liquidity floor, or another token with the same ticker traded more in the last 24 hours. Add your own term with the search box beside the index to bring it in.

    What does the depth check actually compute?

    It treats the pool's quoted liquidity as two equal halves and applies the constant-product formula. A buy of size b against a USD side of L/2 moves the price by roughly (1 + b / (L/2))² − 1. It is a model: real routers split orders across pools, and concentrated-liquidity pools behave differently.

    Does the manifest connect to anything?

    No. It is a document format we propose, built entirely in your browser. The SHA-256 fingerprint lets you publish the file and prove later that it has not changed. Enforcing limits is the job of your session keys, multisig or program — the manifest declares them.

    Where is my watchlist stored?

    In this browser's local storage, under a single key. Clearing site data removes it. Export it as JSON from the watchlist section if you want a copy.

    Do you rank or endorse tokens?

    No. Ordering follows whatever column you sort by. Inclusion means only that a pair matched our filters at the moment you loaded the page. Many agent tokens are thinly traded and can go to zero.

    How current is the framework registry?

    The entries reflect each project's public repository and documentation as our desk last read them. Frameworks change quickly, so confirm licences and features at the source before you depend on them.

    Does Agentfold hold funds or keys?

    Never. There is no wallet connection, nothing to sign and nothing to deposit. The page reads public market data and does arithmetic in your browser.