Overview
PeptAI is a BioAgent: an autonomous research agent launched on Bio Protocol that designs peptide drug candidates computationally, ranks them, and pushes the survivors into physical wet-lab testing. It has its own token, its own treasury, and its own public results feed — but no membership, no proposals and no voters. That makes it structurally unlike every other organisation in this directory, and it is the reason the page exists: PeptAI is the clearest live example of an emerging DeSci shape in which the agent, not a community, is the thing being funded.
Bio Protocol describes the agent as running designed candidates through an eight-gate validation pipeline, publishing each result on-chain so the reasoning behind a candidate can be inspected before anyone spends money synthesising it. Only candidates that clear the gates are sent for synthesis and binding assays. The claim being tested is a narrow one — not that an agent can invent a drug, but that it can cut the cost and latency of the first screening step far enough to make small, community-scale research budgets go further.
What it has actually produced
Two results are worth separating from the marketing around them, because both were reported with a target, a method and a cost.
- OX2R-004 (April 2026). A novel peptide agonist aimed at the orexin-2 receptor, a target of interest in ADHD and narcolepsy research, designed in roughly 24 hours with an initial wet-lab validation cost of about $500–600. The number that matters there is not the 24 hours but the validation cost: it is small enough that a single failed hypothesis is not a budget event.
- VEGFR2 binders (June 2026). Agent-designed binders against VEGFR2, a cancer-relevant receptor, confirmed by surface plasmon resonance — PeptAI's first reported wet-lab confirmation. Bio reports the assay results being fed back into the next design cycle, which is the part that distinguishes this from a one-off screen.
- KISS1R and OX2R candidates (announced 11 August 2026, not yet run). PeptAI confirmed its next two programmes — KISS1R, a fertility target, and OX2R again, for ADHD and sleep — with its agents designing two lead peptides for each. Bio is explicit that the four candidates "so far exist only as predictions on a computer" and are heading to a lab for the first time, to be synthesized and then tested on living cells carrying the target receptor for whether they switch it on, how strongly, and how selectively. Listed here as a pending experiment rather than a result, which is the distinction this page exists to keep. Bio reports on a monthly cadence and the August 2026 edition had not appeared on its blog index when that index was re-read on 14 September 2026 – the newest monthly it lists is July's – so the four candidates remain predictions.
Neither result is a drug. Both are early binding data, the stage at which the great majority of candidates fail, and neither has been independently replicated — the reproducibility problem that DeSci set out to address applies to agent-generated results exactly as it applies to human ones. What is genuinely new is that the hypotheses, the ranking and the outcomes are published on-chain rather than sitting in a private pipeline.
How it was funded: the Ignition Sale
PeptAI did not raise from a token sale in the usual sense, nor from a DAO treasury vote. It was launched through Bio Protocol V2's Launcher, whose Ignition Sale mechanism is worth understanding on its own terms, because it is a genuinely different answer to the raise-allocation problem than either a fixed-price sale or a bonding curve:
- The sale runs in USDC for a fixed window and must hit a stated raise goal to close at all. Miss it, and commitments are returned.
- Participants commit capital and pledge BioXP — points earned by staking BIO or ecosystem tokens, which expire roughly 14 days after they are earned. Anyone can join with no BioXP at a floor weight.
- When a sale is oversubscribed, allocation is pro-rata to pledged BioXP, capped so no single wallet takes more than 0.5% of the tokens on offer. Unused commitments and points are returned.
- On a successful close, the raised USDC is paired with tokens into a liquidity pool rather than being handed over as a lump sum.
PeptAI's sale closed more than 32x oversubscribed, with over 1.6M USDC committed by more than 1,400 participants; PEPTAI then listed on Base on 14 May 2026. Twenty per cent of supply was routed to veBIO stakers, 20% of it claimable at TGE and the remainder vesting over six months, with early full claims forfeiting the unvested portion.
The design is deliberately anti-whale in the same family as the mechanisms catalogued under DAO tokenomics: the scarce good is not capital but time-decaying, stake-earned priority. Whether that survives contact with capital at scale is an open question — points that can be minted by spending BIO are still, at the margin, purchasable.
BioAgent or BioDAO? Why the distinction matters
Bio Protocol now launches three asset types through the same Launcher: IP tokens, BioDAOs and BioAgents. The first BioAgent, Aubrai, directs capital to longevity research; PeptAI is the one that has published wet-lab confirmation. In June 2026 Bio added OpenLabs, a front end where an idea attracts contributors, agents draft the votes and progress summaries, bounties fund the work, and a project can graduate to the Launcher.
For anyone reading this wiki as a governance reference, the important thing is what a BioAgent token does not obviously carry. A VitaDAO-style BioDAO token is a membership claim: it votes on which research to fund and, through Molecule's IP-NFT machinery, on what happens to the resulting intellectual property. Bio's public documentation for the Launcher and its V2 FAQ sets out how allocation works in detail, but does not set out what governance rights an agent's token confers over the agent's treasury, its research direction, or the data and IP it generates.
That gap is the substantive question, not a pedantic one. An agent has no equivalent of low turnout, because there is no electorate to turn out; it also has no equivalent of a member forcing a decision. In a member DAO, holders who dislike the direction can vote, and failing that, leave. Where the funded entity is a model and a lab budget rather than a membership, "exit" reduces to selling on the open market at whatever price is there — which is the same dependency that shows up repeatedly in how DAOs fail. Progressive governance minimisation is a defensible destination for a protocol whose rules are settled; it is a harder argument for an entity whose entire job is to keep making discretionary calls about what to research next.
Where an agent sits on the funding menu
Bio framed the July 2026 monthly around Science: A New Golden Age, the report to the President published that month by Michael Kratsios, Director of the White House Office of Science and Technology Policy. The report's grantmaking chapter is unusually direct about why anyone is experimenting at all — consensus peer review, it argues, “struggles to distinguish the exceptional from the merely good” — and it sets out a menu of replacements: golden tickets, individual-based funding, pull mechanisms that pay for outcomes, scouts rewarded for finding promising work, regranting that delegates allocation “to researchers or experts who possess the specific judgment to identify promising work before consensus forms”, and quadratic funding, which it calls more speculative and still in early testing.
Every entry on that menu decentralises the judgement to more humans. None of them describes what PeptAI is. A BioAgent is not a scout, a regrantor or a matching pool; it is a funded entity that generates its own hypotheses, ranks them, and spends a budget on the survivors, with the humans supplying capital rather than discretion. That is either a seventh mechanism or a category error, and which one it turns out to be is an empirical question the wet-lab results this page is waiting on will start to answer. It is also why the governance gap above matters more here than it would for a grant programme: on the federal menu somebody accountable always chooses, and on this model the chooser has no electorate.
How Caper approaches this
PeptAI is a useful contrast because it isolates one variable: what a funded community's token is actually a claim on. In a caper, that claim is explicit in the contract. A member who wants out calls exit with an amount of governance tokens and an amount of vote tokens, both withdrawn from their own account in the same call, and receives a proportional share of the treasury — no proposal to pass, no majority to win, no counterparty. The vote tokens are the precondition: a caper mints them only for a cast ballot, one per ballot, so a member who has never voted holds none and cannot exit at all. The share is computed as (holdings × votes) / (vote supply × circulating supply), so it reflects both what a member put in and the participation they actually earned; the governance tokens are then sold back into the bonding curve in the same transaction.
That is a narrower claim than "capers do science better" — they do not, and PeptAI's cost-per-hypothesis result is a real contribution this wiki has no equivalent for. The point is only that a member-funded research treasury can be built so that leaving is a contract call rather than a market order, and so that the right to leave does not depend on anyone else agreeing.