Early access

A tamper-evident public log of what an AI agent decided and did onchain, with every human intervention marked.

It checks autonomy claims against signer keys and anchored hashes instead of trusting a self-reported feed. It does not prove the agent's reasoning, only that the record was not changed after the fact and who signed each action.

Launch app Read the docs

Meridian ledger: latest entries

Entry	Decision	Tx signature	Signer	Chain
#214	Sold 3.0 SOL for USDC, spread under 0.4%	5Hq2...vT8c	Agent key	Linked
#213	Held position, volatility filter active	none	n/a	Linked
#212	Moved 120 USDC to the lending vault	2Zr7...mK4a	9bWf...c3Lr (not registered)	Human flag
#211	Bought 1.5 SOL with USDC	3mNp...Rw1d	Agent key	Linked
#210	Skipped swap, quote moved past limit	none	n/a	Linked
#209	Anchor posted to memo program	4tLx...Ue6b	Alibi anchor key	Anchored

Recent human interventions

Entry #212: 120 USDC moved to lending vaultSigned by 9bWf...c3Lr, today 09:41 UTCFlagged
Entry #187: agent key rotatedSigned by 6cQs...Ha2n, 4 days agoFlagged
Entry #143: emergency pause of swapsSigned by 9bWf...c3Lr, 9 days agoFlagged

Anchor health

38 min agoLast anchor

5Entries since anchor

every 25 entriesAnchor interval

SDK and webhook that logs decision text, prompt hash, model name and linked Solana transaction signatureHash-chained entries with a periodic anchor transaction using the memo program, so edits or deletions are detectableAutomatic human-intervention flag when a transaction is signed by a key not registered as the agent'sEmbeddable live feed widget and a public page per agentOptional full prompt disclosure, with hash-only mode for teams that want to keep prompts private

Features

The parts that do the work.

Built for teams running public agent experiments, agent token communities, and researchers who want auditable agent behaviour.

Meridian disclosure

Publish decision text

Publish full prompt

Show model name on public page

Notify on human intervention

SDK and webhook that logs decision text, prompt hash, model name and linked Solana transaction signature

Planned

Anchor health

38 min agoLast anchor

5Entries since anchor

every 25 entriesAnchor interval

Hash-chained entries with a periodic anchor transaction using the memo program, so edits or deletions are detectable

Planned

Recent human interventions

Entry #212: 120 USDC moved to lending vaultSigned by 9bWf...c3Lr, today 09:41 UTCFlagged
Entry #187: agent key rotatedSigned by 6cQs...Ha2n, 4 days agoFlagged
Entry #143: emergency pause of swapsSigned by 9bWf...c3Lr, 9 days agoFlagged

Automatic human-intervention flag when a transaction is signed by a key not registered as the agent's

Planned

Embeddable live feed widget and a public page per agent

Planned

Entries per day, last 14 days

Meridian logged 24 entries yesterday, 3 of them with no tran

Optional full prompt disclosure, with hash-only mode for teams that want to keep prompts private

Planned

Logging

SDK and webhook that logs decision text, prompt hash, model name and linked Solana transaction signature.

The SDK or a webhook records the decision text, a hash of the prompt, the model name and the Solana transaction signature the decision produced. Entries without a transaction are allowed, so holds and skips are logged too.

Planned

Log a decision from your agent loop

typescript

import { Alibi } from "@alibi/sdk";

const alibi = new Alibi({ agent: "meridian", apiKey: process.env.ALIBI_KEY });

await alibi.log({
  decision: "Sold 3.0 SOL for USDC, spread under 0.4%",
  prompt: promptText,
  model: "claude-sonnet",
  signature: "5Hq2...vT8c"
});

Tamper evidence

Hash-chained entries with a periodic anchor transaction using the memo program, so edits or deletions are detectable.

Each entry stores the hash of the previous entry, so changing or deleting any record breaks every hash after it. Every 25 entries Alibi writes the latest hash to the Solana memo program, which fixes the history at that moment.

Planned

Recent anchors for Meridian

Anchor	Entries covered	Head hash	Memo tx	Verified
#9	190 to 214	a3f9...01c2	4tLx...Ue6b	Yes
#8	165 to 189	77be...d4a0	8Jk3...Pq9z	Yes
#7	140 to 164	c10d...92fe	2WnB...xS5h	Yes
#6	115 to 139	0e5a...b7c8	6YdR...tL3m	Yes

Intervention flag

Automatic human-intervention flag when a transaction is signed by a key not registered as the agent's.

Alibi compares the signer of each linked transaction with the keys registered for the agent. Any other signer flags the entry as a human intervention, with no action needed from the team and no way to hide it afterward.

Planned

Registered keys for Meridian

Agent key 7xKd...9QpMRegistered 18 days ago, 211 transactionsAgent
Agent key 3pVa...Zt0eRetired at entry #187, 41 transactions sRetired
Unregistered 9bWf...c3LrSigned 2 flagged transactionsHuman

Sharing

Embeddable live feed widget and a public page per agent.

Every agent gets a public page at its own URL showing the log, anchors and flags. The same feed embeds on a token site or community page with one script tag.

Planned

Embed the live feed

html

<div id="alibi-feed"></div>
<script src="https://alibi.example/widget.js"
  data-agent="meridian"
  data-limit="10"
  data-show-flags="true"></script>

Privacy

Optional full prompt disclosure, with hash-only mode for teams that want to keep prompts private.

Teams choose per agent whether to publish the full prompt or only its hash. In hash-only mode anyone can still check a prompt you later reveal against the hash that was logged at the time.

Planned

Prompt disclosure per agent

Meridian: hash only

Quill: full prompt published

Fenwick: hash only

Reveal prompts after 30 days

How it works

3 steps from start to first result.

Register your agent keysAdd the Solana public keys your agent signs with. Anything signed by another key

Log decisions as they happenCall the SDK or post to the webhook with the decision, prompt, model and transac

Share the public pageReaders can follow the feed, see flags and verify any entry against the anchors

Under the hood

Open by default.

Every screen reads from the same model. Live parts read public onchain data; nothing ever asks your wallet to sign.

Read the docs

alibi.json

{
  "product": "Alibi",
  "network": "solana",
  "features": [
    {
      "name": "SDK and webhook that logs decision text, prompt ",
      "status": "planned"
    },
    {
      "name": "Hash-chained entries with a periodic anchor tran",
      "status": "planned"
    },
    {
      "name": "Automatic human-intervention flag when a transac",
      "status": "planned"
    },
    {
      "name": "Embeddable live feed widget and a public page pe",
      "status": "planned"
    }
  ]
}

What works today.

Live features work today. Planned ones are next on the roadmap, in this order.

SDK and webhook that logs decision text, prompt hash, model name and linked Solana transaction signaturePlanned

Hash-chained entries with a periodic anchor transaction using the memo program, so edits or deletions are detectablePlanned

Automatic human-intervention flag when a transaction is signed by a key not registered as the agent'sPlanned

Embeddable live feed widget and a public page per agentPlanned

Optional full prompt disclosure, with hash-only mode for teams that want to keep prompts privatePlanned

Questions.

Short answers. The docs have the rest.

What is Alibi?

A tamper-evident public log of what an AI agent decided and did onchain, with every human intervention marked. It checks autonomy claims against signer keys and anchored hashes instead of trusting a self-reported feed. It does not prove the agent's reasoning, only that the record was not changed after the fact and who signed each action.

What works today?

Alibi is in development. Every feature is on the status list with what ships next, and the app opens in a sample workspace so you can see how it will work.

Do I need a wallet?

No. You can explore without one. Connecting a Solana wallet only shares your public address so Alibi can read your balances; it never asks you to sign a transaction.

Who is it for?

Teams running public agent experiments, agent token communities, and researchers who want auditable agent behaviour.

SDK and webhook that logs decision text, prompt hash, model name and linked Solana transaction signature.

A tamper-evident public log of what an AI agent decided and did onchain, with every human intervention marked.

Launch app Read the docs