Platform guides

Decentralized OSINT: what persists on IPFS, Arweave, Odysee, Nostr and the Fediverse.

Five networks, five different answers to the same question: when something is deleted, who still has it, and how do you find it? This page gives the persistence rule, the working search tool and the evidence method for each, verified 9 September 2026.

Short answer

Search each network by its own identifier: CIDs through cid.contact and public gateways for IPFS (there is no working keyword engine), wallet and tag queries on arweave.net/graphql for Arweave, claim IDs and yt-dlp for Odysee, npubs through noogle.lol for Nostr, and hashtags and account URLs for Mastodon. For the Reddit side, where these links were posted and later deleted, THINKPOL holds 30 billion Reddit records including deleted content: search decentralized links across the Reddit archive, free, no account.
01

What is decentralized social media, and why does it change the deletion question?

Decentralized social media is any platform where no single company holds the content and the accounts: files addressed by hash on IPFS, data written permanently to the Arweave blockchain, video metadata on the LBRY chain behind Odysee, signed notes relayed across Nostr, and federated servers running Mastodon and the rest of the Fediverse. Last verified: 9 September 2026. On a centralized platform, deletion is one decision by one operator. On these networks deletion is a question about who else kept a copy, and the answer differs for every protocol on this page.

Two disambiguations before anything else, because the keyword data is polluted by them. "IPFS" in most US searches means IPFS Corporation, an insurance premium finance company; this page is about the InterPlanetary File System. "LBRY" in search data is largely a typo for "library"; here it means the LBRY protocol and blockchain that Odysee runs on.

02

Can IPFS content be deleted, and how do you search IPFS in 2026?

IPFS content can disappear, but nobody can delete it for you. A file on IPFS is addressed by its content identifier (CID), a hash of the content itself. It stays reachable as long as at least one node holds it and announces it; nodes run garbage collection on unpinned data, and the IPFS documentation is explicit that the network "doesn't guarantee that any content is persistently available." Pinning, on your own node or through a pinning service, is what keeps a CID alive. So a takedown against IPFS means persuading every pinner to unpin, which is why material that was pinned by several parties tends to outlive the request, and material nobody pinned vanishes on its own.

Search is the weak point, and Google's own AI Overview says so: keyword search on IPFS is "difficult." The general-purpose engine ipfs-search.com suspended its servers on 31 May 2023 for lack of funding; on 9 September 2026 its domain shows only a "for sale" page. A newer indexer at ipfssearch.com did not respond to our requests that day. What works is lookup by CID rather than by keyword. The Network Indexer at cid.contact (online) tells you which providers currently announce a CID. Any public gateway (ipfs.io, dweb.link, Cloudflare's) serves the content if a provider has it. The IPLD explorer at explore.ipld.io (online) walks the structure behind a CID, which is how you find the files inside a directory hash without a keyword index. Verdicts: cid.contact and gateways, use them; ipfs-search.com, dead; keyword search across IPFS, not available from any public service we could reach.

03

Arweave and the permaweb: ViewBlock, ArScan and the GraphQL gateway

Arweave is the opposite design: data is paid for once and written to a blockchain designed to keep it for at least 200 years. The "permaweb" is the set of web content and applications stored that way. There is no delete operation. A transaction on Arweave stays readable at arweave.net/TRANSACTION_ID, and the only way it becomes unreadable is if every gateway declines to serve it, which does not remove it from the chain.

Three tools were working on 9 September 2026. ViewBlock's Arweave explorer (viewblock.io/arweave) resolves transactions, wallets and blocks with a familiar block-explorer interface. ArScan (arscan.io) does the same with better tag filtering. The GraphQL gateway at arweave.net/graphql answered a live query from us that day and is the investigator's tool: every Arweave transaction carries tags (Content-Type, App-Name, and whatever the uploading application added), and a GraphQL query can return every transaction from a wallet, every upload by an application, or every item tagged with a given name, in reverse chronological order. That makes a wallet address a stable identifier across years of uploads. Verdicts: all three working; the GraphQL gateway is the one that scales.

04

Odysee and LBRY: what stays on the chain when a channel is deleted

Odysee is a video platform built on the LBRY protocol. Channel and video metadata (the "claims": names, descriptions, thumbnails, the publishing channel's key) are written to the LBRY blockchain; the video files themselves are hosted by Odysee's infrastructure and a network of hosts, not on the chain. Corporate history matters for who answers a request today: Odysee split from LBRY Inc. on 1 October 2021, LBRY Inc. lost its SEC case in July 2023 and wound down, and Odysee was acquired by Forward Research in June 2024. Odysee.com and lbry.com were both online on 9 September 2026.

Deletion on Odysee removes the hosted video and marks the claim as abandoned on the chain. The claim's history (its name, its channel, its timestamps and the transaction that created it) remains readable in the blockchain, so a deleted channel is still provable as having existed and having published a given title on a given date, even though the video is gone. That is the investigative shape of Odysee: the chain proves publication; only a copy made in time proves content. For preservation, yt-dlp lists odysee.com, its channels and its playlists as supported sources, and running it with the download-archive option keeps a ledger of what was captured. The "odysee downloader" sites that rank for that query are browser wrappers around the same idea, with adverts. Verdicts: the chain, use it for publication history; yt-dlp, use it for evidence capture; downloader sites, unnecessary.

05

Nostr: relays, npubs and search

Nostr has no accounts and no servers of record: a user is a keypair (the public key shown as an npub), every note is signed by it, and notes are pushed to and pulled from relays, which are simple servers anyone can run. Deletion is a request, not an action. A user can publish a deletion event, and well-behaved relays honour it, but any relay or client that kept the note still has it, and nothing forces them to comply. Notes you have already fetched remain valid signed evidence regardless of what happens on the relays afterwards.

Search depends on the relay. NIP-50 defines a search filter that relays may support, and the specification itself tells clients to check each relay's supported NIPs and to query several relays because coverage differs. In practice you search through an aggregator: noogle.lol ("Nostr decentralized search") was online on 9 September 2026; nostr.band, the other aggregator commonly cited, refused connections that day. An npub is a stable identifier across every relay and client, which makes it the pivot for identity work, in the same way a Reddit username is in the Reddit username guide.

06

Mastodon and the Fediverse: instance search, full-text limits and defederation

The Fediverse is thousands of independent servers (instances) speaking ActivityPub, with Mastodon the largest software. Content lives on the author's instance and is copied to every instance that has a follower of the author. Deletion propagates as a request to those copies, and most honour it, but an instance that is offline, that has defederated from the author's server, or that simply ignores the request keeps its copy. Bluesky, which people often include here, runs on a different protocol (AT Protocol), not ActivityPub, and is bridged to the Fediverse only through third-party services.

Search is deliberately limited. Mastodon's documentation states that full-text search is optional for admins to install and, where installed, is restricted for safety reasons to your own posts, favourites, bookmarks and mentions; there is no network-wide keyword search by design. Hashtag search works, and pasting a remote account's or post's full URL into any instance's search box fetches it. Instance-level search of public posts exists on some servers that opted in, and research tools such as Indiana University's OSoMe Mastodon search index public posts from a set of large instances. The investigator's practical path is hashtags, account URLs, and the instance's public local timeline, plus the fact that an account name is user@instance and the instance's domain is itself a lead.

NetworkWhat persistsCan it be deletedHow you search itEvidence method
IPFSAny pinned file, by CIDOnly if every pinner unpins; unpinned data is garbage-collectedBy CID: cid.contact, gateways, explore.ipld.io; no working keyword engineFetch by CID, record the CID and the gateway response
Arweave / permawebEverything written, permanentlyNoviewblock.io/arweave, arscan.io, arweave.net/graphql by wallet or tagCite the transaction ID; it does not change
Odysee / LBRYClaim metadata on chain; video hosted off chainVideo yes; claim history noOdysee search, LBRY chain explorers, yt-dlp for captureCapture the video with yt-dlp, cite the claim ID
NostrSigned notes on whichever relays kept themDeletion is a request relays may ignorenoogle.lol; NIP-50 relays; by npubSave the signed event JSON; the signature is the proof
Mastodon / FediverseCopies on every instance that follows the authorPropagated request; offline or defederated instances keep copiesHashtags, account URLs, local timelines; no global full-text searchScreenshot plus the post URL on the home instance
07

The decentralized trail on Reddit

Every one of these networks was discussed, promoted and linked on Reddit before it had a search engine of its own, and the archive shows the cycle. Reddit posts and comments mentioning "odysee" in THINKPOL's archive peaked at 38,200 in 2021 and fell to 11,800 in 2025; "ipfs" peaked at 19,700 in 2021 and fell to 4,300 in 2025; "arweave" peaked at 4,500 in 2022 and stood at 1,900 in 2025; "lbry" peaked at 10,300 in 2021 and stood at 1,300 in 2025. Inside those counts are the links that matter: an Odysee channel URL posted in a subreddit that was later banned, an IPFS hash shared as a "backup" of a removed document, a permaweb link to a manifesto. When the subreddit goes and the Reddit post goes, the archive still holds the link, the account that posted it and the date.

Try it without an account: search Reddit for odysee.com links or for ipfs.io/ipfs/ hashes, then pivot on the posting account with the Reddit user lookup. For a community that was banned and rebuilt itself elsewhere, the pattern is documented in our respawn network investigation, and the banned community itself can be reconstructed with the subreddit lookup.

08

What THINKPOL adds today

THINKPOL does not index IPFS, Arweave, Odysee, Nostr or the Fediverse. It archives Reddit: 30 billion posts and comments since 2005, captured at publication and retained after deletion, full-text searchable in under 300 milliseconds, hosted in the EU. For a decentralized-platform case that is the discovery layer: the links, hashes, channel names and wallet addresses that were posted on Reddit, including in posts and communities that no longer exist, and the Reddit identities behind them. Mastodon and the Fediverse are on the roadmap behind Telegram and Discord. The free tools need no account; the wider context of where this sits is on the grey web page.

Need the full archive via API?

The free tools query a slice of the archive. The API gives you all 30 billion posts and comments back to 2005, deleted content included, in under 300ms.

FAQ

Decentralized platforms, answered.

Platforms where no single company holds the content and accounts: content-addressed files on IPFS, permanent data on the Arweave blockchain, video metadata on the LBRY chain behind Odysee, signed notes on Nostr relays, and federated servers running Mastodon and the wider Fediverse. Deletion on each one depends on who else kept a copy.

Not by anyone but the nodes holding it. A file stays reachable while at least one node pins and announces its CID; unpinned data is garbage-collected. A takedown means persuading every pinner to unpin, which is why widely pinned material persists and unpinned material fades on its own.

Not a working general one on 9 September 2026. ipfs-search.com shut down on 31 May 2023 and its domain is for sale; ipfssearch.com did not respond to our requests. Lookup by CID works through cid.contact, public gateways and the IPLD explorer.

Odysee is a functioning video site owned by Forward Research since June 2024. The risks are content-side: lightly moderated material, and video files hosted off the blockchain that can be removed while the claim metadata remains. Treat downloads and links there as you would on any unmoderated platform.

Forward Research, which acquired Odysee in June 2024. Odysee had split from LBRY Inc. on 1 October 2021; LBRY Inc. lost its SEC case in July 2023 and wound down. The LBRY blockchain the platform runs on continues independently.

No. Bluesky runs on the AT Protocol, not ActivityPub, so it is not natively federated with Mastodon and the Fediverse. Third-party bridges connect the two, but accounts and posts are not shared by default.

Only partially. Hashtag search works everywhere and pasting an account or post URL fetches it. Full-text search is optional per instance and, where enabled, limited to your own posts, favourites, bookmarks and mentions. There is no network-wide keyword search by design.

The set of web pages and applications stored permanently on the Arweave blockchain. Data is paid for once and cannot be deleted; a transaction stays readable at its ID through gateways such as arweave.net, and explorers such as ViewBlock and ArScan index it.

Keep reading

Telegram OSINT

The platform where most of these links get shared first, and what persists there.

What is the grey web?

Where decentralized and semi-public platforms sit between the open and dark web.

OSINT tools landscape

45 vendors across 10 categories, mapped by what they actually cover.

THINKPOL is an independent intelligence platform and is not affiliated with, endorsed by, or sponsored by Reddit Inc., Telegram, Discord Inc., Meta Platforms, Tencent, LY Corporation, or any other platform or third-party tool named on this page. Platform names are trademarks of their respective owners. Third-party tool descriptions reflect publicly observable functionality on the verification date stated on the page and may change; corrections are welcome via our contact page.

Stop reading about
it in the news.

Request access to THINKPOL. We respond within one working day. A 30-minute scoping call follows, and a sandbox tenant is provisioned within five working days of contract signature.

Contractual agreement requiredSandbox in 5 working days
© 2026 THINKPOL SAS
Backed byFrance 2030APOK InvestLa French Tech