# SIMD / Identity.md: AI Agent Swarm Research Report

**Research question:** What is Identity.md (IMD), how does it work, what does it cost to use or join, can participants earn from it, and what are the risks? The report also asks how "SIMD" relates to it.

**Snapshot time:** 2026-10-06 ~05:17 UTC. All live figures below were read from public endpoints at that time and will change.
**Method:** I read the official docs, called the public control-plane API (`api.imd.fun`) directly, viewed the public explorer, and read one independent technical writeup plus two secondary sources. I made no paid requests, connected no wallet, and did not run a worker. Nothing in this report is financial advice.

## How to read this report

Each claim carries one of these labels:

| Label | Meaning |
|---|---|
| **[FACT-P]** | Stated in a primary source (official docs or the live API response), with a citation |
| **[FACT-3P]** | Reported by a third party that I could not independently confirm |
| **[INFERENCE]** | My own reasoning from the cited facts |
| **[UNCERTAIN]** | Sources conflict, or the data is incomplete or ambiguous |
| **[OPEN]** | Unanswered question |

Source keys such as [S1] are listed at the end.

---

## 1. Overview

- **[FACT-P]** The docs describe IMD as a network where users pay for **jobs** (code, contracts, websites, research), **launches** (jobs that end in a contract deployment), **workflows** (contract → review → deploy → website), **oracle questions** answered by agent panels, and **schedules** of recurring runs. [S1]
- **[FACT-P]** Workers ("agents") are tied to **seats**, which are NFTs in the collection `0x0000ec93127baa929e58e97dd0095a2bfb38ec1d` on Ethereum mainnet (chainId 1). The live `/health` endpoint reports this collection as the identity collection. [S4]
- **[FACT-3P]** The seat collection holds 2,000 NFTs and is fully minted, so new operators must buy a seat on the secondary market. [S3b][S6] I did not check the supply on-chain.
- **[FACT-P]** The protocol reports `protocolVersion: 1` and `deployedAt: null`. [S4 `/version`]

### What "SIMD" refers to — [UNCERTAIN]
The official docs do not define "SIMD". [S1] I found two unrelated things that use the name:
1. **[FACT-3P]** **superimdc.xyz** describes "SIMD — Super Intelligent Identity", an ERC-20 token on Robinhood Chain (4663) at `0x577bc74a99aabbd43a354668dc7b7ed4b5e303ae`. It claims to use "protocol fees to cover 100% of Identity.md job costs". The site names no operator, and I found no evidence that the IMD team endorses it. [S7]
2. **[FACT-3P]** Bankless says $IMD can be staked for "sIMD yield-bearing shares". [S6]

**[OPEN]** Which of these the task means by "SIMD" is unknown. I have not verified that the SIMD token pays for Identity.md jobs. Treat it as a separate, third-party project unless the IMD team says otherwise.

## 2. How the platform works

- **[FACT-P]** **Requesting work:** `POST /requests/quote` → `POST /requests/:id/submit`, which carries an x402 payment challenge, a Permit2 authorization, and EIP-712 quote approval. The client then polls until the request is admitted. Quotes expire after 600 s. [S1][S4 `/requests/capabilities`]
- **[FACT-P]** **Working:** a seat holder pairs an Ed25519 device key to the seat (`/pair/start`, `/pair/complete`) and binds the seat to an ERC-8004 agent. A daemon then claims work over a WebSocket. [S1]
- **[FACT-P]** **Verification:** the docs say submissions produce artifacts that are judged against acceptance criteria. Oracle panel members sign EIP-712 attestations that contracts can verify. Reviews and work records are published at `/reviews/:hash.json` and `/work-records/:hash.json`. [S1] The independent writeup names the judges as "verifier-rerun" and "verifier-paths" for each skill. [S3a]
- **[FACT-P]** **Transparency:** jobs, seats, workers, launches and earnings can all be read through public API routes such as `/jobs`, `/seats/:tokenId`, `/workers`, `/launches/:id?claims=1` and `/wallets/:address/earnings`. The explorer offers a UI over the same data. [S1][S2]
- **[INFERENCE]** Anyone can check what the network *says* it did. "Accepted" still only means the platform's own verifier passed the work. No outside party audits the quality of accepted work.

## 3. Costs and requirements

### For requesters
- **[FACT-P]** Each paid action costs **0.5 IMD**: `job.open`, `job.continue`, `launch.open`, `workflow.open` and `oracle.request`. `schedule.create` and `schedule.topup` cost 0.5 IMD per run. [S1] The live capabilities endpoint confirms `amount: 500000000000000000` (18 decimals) for every action. The asset is `0xd34a99bc0f67ae1bbd63c660e6d0b0dd03e263b7` on `eip155:1`, and the recipient (payTo) is `0x4e0fa57bde726079356537e2f34d671e9f41adbc`. [S4]
- **[FACT-P]** The server's wallet pays the gas for the payment transaction. [S1]
- **[FACT-P]** For schedules, the docs say "Skipped and failed runs cost nothing". A schedule cancelled after the quote but before payment confirms is refused, and "the runs are not refunded". [S1]
- **[INFERENCE]** Before approving a request, confirm three things. The token contract should match `0xd34a…63b7`. The network should be Ethereum mainnet. The amount should equal what `/requests/capabilities` returns. The price is set by the server and could change.

### For worker operators
- **[FACT-P]** You need a seat NFT, a paired device key, and an ERC-8004 agent registration. [S1]
- **[FACT-3P]** The independent setup guide [S3b] lists:
  - an AI CLI subscription (Codex CLI or Claude Code), whose quota you pay for;
  - a Linux machine that stays online, about 2 vCPU / 4 GB RAM / 40 GB disk, estimated at roughly $20/month;
  - Foundry, for contract skills;
  - a small amount of mainnet ETH for the one-time ERC-8004 registration transaction.

  The same guide reports that a ChatGPT Plus quota ran out in about 7 minutes during a burst of tasks. Bankless also mentions the Claude/Codex subscription requirement. [S6]
- **[UNCERTAIN]** Bankless gave a seat floor price of about 1.99 ETH on 2026-09-25 [S6]. I did not check the current price.

## 4. Potential earning opportunities

- **[FACT-P]** Workers are paid in **launch tokens**, not IMD and not cash. Each launch gives 10% of its token supply to the swarm and 90% to the requester. The swarm share is claimed from a Merkle distributor: "2% split equally between the wallets with accepted work on the launch, 8% equally per seat connected at admission." [S1] Claims can be checked at `/launches/:id?claims=1` and `/wallets/:address/earnings`. [S1]
- **[UNCERTAIN]** The independent guide describes the 8% differently: "split equally among wallets with accepted work in the preceding 12 h". [S3b] The official docs say "per seat connected at admission". **I treat the official docs as authoritative.** The rule may have changed between versions.
- **[FACT-P]** Live launch data on 2026-10-06 shows 248 launches in total [S4 `/launches`]:

  | Chain | live | parked | abandoned |
  |---|---|---|---|
  | Sepolia testnet (11155111) | 177 | 41 | — |
  | Ethereum mainnet (1) | 17 | 1 | 10 |
  | Robinhood Chain (4663) | 2 | — | — |

  The default launch chain is still Sepolia, and mainnet and Robinhood Chain are now also offered. [S4 `/requests/capabilities`]
- **[INFERENCE]** Most allocations so far are testnet tokens. Testnet tokens have no market value by design. A mainnet allocation only has value if that token has liquidity, and I did not check the liquidity of any launch token.
- **[FACT-3P]** On 2026-09-24 the guide said all 10 mainnet launches were abandoned. [S3b] That is no longer true: 17 mainnet launches are now live. The writeup's other figures (e.g. "100 jobs") are also out of date.
- **[INFERENCE]** Earnings depend on four things: owning a seat, keeping a worker online at the right time, getting work accepted, and the market value of each launch token. None of this is guaranteed, and the operator always pays for the AI subscription and the server. **No profit should be assumed.**

## 5. Live activity (snapshot 2026-10-06 ~05:17 UTC)

| Metric | Value | Source |
|---|---|---|
| Connected daemons / agents online | 643 | [S4 `/health`, `/swarm`] |
| Active enrollments (seats) | 652 | [S4] |
| Accepted steps, last 24 h | 44,000 | [S4] |
| Jobs done / oracle answers, last 24 h | 329 / 312 | [S4 `/swarm`] |
| Total jobs (API) | 8,799 (8,610 completed, 126 cancelled, 59 blocked, 4 executing) | [S4 `/swarm`] |
| Total jobs (explorer page) | 1,306 (1,219 completed) | [S2] |
| Payment orders | 1,264 paid, 20 failed (19 `payment_permission_expired`), 1,214 expired | [S4 `/health`] |
| Verifier / publisher / deployer | all up | [S4] |

- **[UNCERTAIN]** The explorer reports 1,306 jobs and the API reports 8,799. The explorer probably counts a subset, such as non-oracle jobs, but I could not confirm this. Use the API figures with care, and treat any single dashboard number as possibly incomplete.
- **[INFERENCE]** Almost as many quotes expired (1,214) as were paid (1,264). An expired quote is not a lost payment. It still suggests that many people start the payment flow and do not finish it, which the writeup also noted. [S3a]
- **[INFERENCE]** Online seats grew from 263 (writeup, Sep 24) [S3a] and 372 (Bankless, Sep 25) [S6] to 643 today. More seats competing for a fixed 8% share means a smaller slice per seat.

## 6. Risks and limitations

- **Token prices and liquidity may fluctuate.** [FACT-3P] Bankless cited about $9.73 per IMD and a market cap of about $69M on 2026-09-25 [S6]. KuCoin's headline mentioned an all-time high of about $39M market cap [S8]. These figures disagree in time and size, and I did not check the current price.
- **Operators pay costs whatever they earn.** Running an agent means paying for an AI subscription and a server. [FACT-3P, S3b]
- **Dashboard metrics may be unavailable or incomplete.** Job counts differ between the explorer and the API (section 5). [FACT-P]
- **A successful check does not guarantee a reward.** Rewards come from launches. A job that is not a launch, or a launch on testnet, produces no tradable reward. [INFERENCE from S1, S4]
- **The project is young and unaudited.** It runs protocol v1, and Bankless advises treating it "like an experiment that hasn't been formally audited". [FACT-P S4; FACT-3P S6]
- **Auto-updates mean trusting the release pipeline.** The guide says the worker auto-updates from GitHub releases, so operators run whatever the release pipeline ships. [FACT-3P, S3b]
- **Look-alike projects.** Projects with similar names, such as SIMD, exist and are not shown to be affiliated (section 1). Verify every contract address against the official docs. [INFERENCE]

## 7. Unanswered questions

1. What does "SIMD" mean in this task, and does the SIMD token actually pay for Identity.md jobs? (Not verified.)
2. What is the current market price and liquidity of IMD and of mainnet launch tokens? (Not checked.)
3. What does a typical seat earn per day, in realised value? There is no public aggregate. Per-wallet data exists at `/wallets/:address/earnings`, but I did not sample it.
4. Why do the explorer and API job counts differ?
5. Has any outside party audited the contracts (token, seat collection, Merkle distributors, burn hook)? `/launches/:id/assurances` exists per launch [S1], but I did not survey it.
6. Which 8% distribution rule applies today: the official docs or the independent guide's description?

## 8. Conclusion

**[FACT-P]** Identity.md is a live, actively used system. It runs a public, machine-readable control plane on Ethereum mainnet and charges a fixed 0.5 IMD per paid request. Hundreds of seats are online and tens of thousands of steps are accepted per day. [S1][S4]

**[INFERENCE]** Its practical value depends on four things this report could not verify:
- how good accepted work really is;
- how robust the platform's own verifiers are;
- how much paid demand comes from real users rather than test runs;
- whether launch-token rewards keep a market value that covers the subscription and server costs operators pay.

Beginners should start with read-only research using the public API and explorer. They should not spend IMD or buy a seat until they understand the payment flow, the testnet-versus-mainnet distinction, and the risks above.

---

## Sources

- **[S1]** Project documentation — https://www.imd.fun/docs/ (fetched 2026-10-06; primary)
- **[S2]** Public job explorer — https://explorer.imd.fun/ (viewed 2026-10-06; primary)
- **[S3a]** mtezy, *IdentityMD (IMD) — Writeup*, snapshot 2026-09-24 — https://github.com/mtezy/imd-writeup (independent, third party)
- **[S3b]** mtezy, *Worker Setup Guide*, snapshot 2026-09-24 — https://github.com/mtezy/imd-writeup/blob/master/WORKER-SETUP.md (independent, third party)
- **[S4]** Live control-plane API, read 2026-10-06 ~05:17 UTC — https://api.imd.fun/version, /health, /swarm, /requests/capabilities, /launches?limit=500 (primary, time-sensitive)
- **[S6]** W. M. Peaster, "Inside IMD, Ethereum's New AI Swarm Experiment", Bankless, 2026-09-25 — https://www.bankless.com/read/inside-imd-ethereum-s-new-ai-swarm-experiment (secondary)
- **[S7]** "SIMD — Super Intelligent Identity" — https://www.superimdc.xyz/ (third party; affiliation unverified)
- **[S8]** KuCoin blog, "What Is IMD Token?" — https://www.kucoin.com/blog/imd-token-community-owned-ai-agents (secondary; headline only, not read in full)

The report keeps the project's own described features [FACT-P] separate from third-party claims and from potential benefits that still need independent verification.