
OrcaID: An Account in Your Agent's Name, Not Yours
- DeepSeekNEWDeepSeek: DeepSeek V4 Flash Vision (Exp)2026-08-21$0.15 / $0.29 per 1M tokens
- z-aiNEWZ.ai: GLM 5.32026-08-1860Intelligence75Coding
- obsidianNEWQwen3.8 27B2026-08-1552Intelligence68Coding
- qwenNEWQwen: Qwen3.8 27B (free)2026-08-13qwen/qwen3.8-27b-free
- deepseekNEWDeepSeek: DeepSeek V4 Pro 08132026-08-1253Intelligence69Coding
- grokNEWSpaceXAI: Grok 4.62026-08-1261Intelligence77Coding
- metaMeta: Muse Spark 1.22026-08-0557Intelligence72Coding
- qwenQwen: Qwen3.8 Max2026-08-0358Intelligence72Coding
- deepseekDeepSeek: DeepSeek V4 Flash 07312026-07-3152Intelligence69Coding
- minimaxMiniMax: MiniMax-H32026-07-31minimax/minimax-h3
- qwenQwen: Qwen3.7 Flash2026-07-27$0.03 / $0.13 per 1M tokens
- orcaOrcaDub: OrcaDub 1.02026-07-27orca/dub
- anthropicAnthropic: Claude Opus 52026-07-2463Intelligence78Coding
- googleGoogle: Gemini 3.6 Flash2026-07-2152Intelligence69Coding
- googleGoogle: Gemini 3.5 Flash-Lite2026-07-2137Intelligence49Coding
- metaMeta: Muse Spark 1.12026-07-1653Intelligence71Coding
- kimiMoonshotAI: Kimi K32026-07-1560Intelligence76Coding
- openaiOpenAI: GPT-5.6 Luna2026-07-0952Intelligence71Coding
- openaiOpenAI: GPT-5.6 Terra2026-07-0957Intelligence77Coding
- openaiOpenAI: GPT-5.6 Sol2026-07-0961Intelligence77Coding
Every autonomous agent shipping today has the same hole in it: when it acts, it acts as you. It sends mail from your address, pays with your card, and signs into your accounts with your session. OrcaID is a bet that this is the wrong shape — that an agent should have an account in its own name, with a wallet, a card, an inbox and a phone number issued to it and governed by you. It is in pre-registration right now: you can reserve a handle, and reserving does not open an account or start billing.
The reason this matters in August 2026 rather than a year ago is that agents stopped being chat windows. SpaceXAI launched Grok Bot on 11 August as an "always-on" teammate that runs on a cloud computer and keeps working after you close the laptop. OpenClaw has gathered roughly 355,000 GitHub stars in five months running on people's own machines with access to their files and mail. Hermes Agent from Nous Research is designed to run for months, writing its own skills as it goes. All three are useful precisely because they act without you watching. And all three, at the moment of acting, are wearing your identity.
What OrcaID actually is
The pitch on orcaid.ai is one sentence: an account in your agent's own name. "Wallet. Card. Inbox. Number. Issued to your agent, governed by you."
The unit is a handle. You reserve @name, and it resolves as name.orcaid.ai — lowercase letters, digits and inner hyphens, up to 63 characters, described on the site as "one of one". Everything else derives from that name. The site's own three-step framing is Reserve ("the name is the address"), Fund and cap ("a balance and a ceiling it cannot argue past"), and Pull ("pause it and everything goes dead").
Four capabilities hang off the handle. It is worth being precise about their status, because the site is:
Wallet. Status on orcaid.ai: By verification. What it is for: A stablecoin balance in the agent's name. Every call, charge, message and minute draws from it.
Card. Status on orcaid.ai: By verification. What it is for: Virtual and merchant-locked, cleared against that balance in the moment.
Inbox. Status on orcaid.ai: At launch. What it is for: The whole domain is the agent's — any address on it, scanned before anything reaches the model.
Number. Status on orcaid.ai: By verification. What it is for: Voice and SMS, for when the agent has to reach a person. Always recorded, always transcribed.
Those labels — "By verification", "At launch" — are the site's own, and they mean what they say. This is a reservation stage, not a shipped product. What you can do today is hold a name.

Why a wallet is not enough
There is a small industry building agent wallets — give the thing a balance and let it spend. OrcaID's framing draws a sharper line, and it is the most useful idea on the page: "A wallet lets an agent spend. An account lets it act."
Spending is only one of the things an agent has to do to finish a real task. Book a venue and someone emails back a contract. Sign up for an API and a verification code arrives by SMS. Dispute a charge and a human wants to talk on the phone. An agent with a balance and nothing else stalls at every one of those steps, and the workaround is always the same: it borrows the operator's inbox, the operator's number, the operator's card. The borrowing is the problem, not the paying.
That is why the four capabilities are packaged together rather than sold separately. A handle that owns a domain's worth of email addresses, a phone line, a card and a balance can complete a task end to end without ever touching your identity. And because all four derive from one name, there is exactly one thing to fund, one thing to cap, and one thing to pull.
The part that makes it governable
The capability list is the easy half. The interesting half is what the operator gets.
The dashboard shown on the site has three controls: a monthly cap described as a hard stop, a kill switch that is one click to pause, and a statement that puts "tokens + dollars, one ledger". The principal — you — is a named field on the account.
Read those together and the design intent is clear. The agent's authority is bounded by a number it cannot argue past, revocable in one action, and legible after the fact in a single ledger that mixes model spend and real-world spend. Compare that to the status quo, where an always-on agent's authority is bounded by whatever your saved card and your logged-in browser session happen to permit, revocable only by changing your own passwords, and legible only if you go reconstruct it from four different billing portals.
The wallet detail worth pulling out is "funds reserve before they're spent. Nothing clears twice." That is an idempotency guarantee, and it is aimed squarely at the failure mode of an unattended agent: a task that retries, and pays twice.

Where it sits relative to the agents you already run
OrcaID is not an agent and does not compete with one. It is the account layer underneath, and the three agents in the news each show a different reason to want one.
Grok Bot is the sharpest case. SpaceXAI's own description says the bots sign into the tools you already use. The most common criticism from analysts covering the 11 August launch was not about capability — it was that the agents log in as you, and SpaceXAI has not published how the authentication works. An agent that has its own account does not need to log in as anyone.
OpenClaw shows the cost of the current arrangement. It runs on your machine, reads and sends your email, and executes code. Its security record in 2026 includes more than 135,000 exposed instances, CVE-2026-25253 at CVSS 8.8, and a supply-chain campaign — all reported by third parties, not the project. When the agent's blast radius is your identity, an exposed instance is an exposed you.
Hermes Agent shows the workaround becoming official. Nous Research's build is model-agnostic, runs for months, and in v2026.8.16 shipped Bitwarden and 1Password integration. An agent reaching into a human password vault is a reasonable engineering decision and also a plain statement that it has no credentials of its own.
None of these is an OrcaID integration. No connector to any of them has been announced, and nothing here should be read as one. The point is narrower: three of the most-discussed agent projects of 2026 have independently arrived at the same missing primitive.

The takeaway
OrcaID is a reservation right now, and it should be judged as one. The wallet, card and number are marked "by verification" and the inbox "at launch" on the company's own page; the only thing you can do today is hold a handle, free and without a card. Whether the execution lands is a question for the launch, not for this article.
What the idea gets right is the diagnosis. Every serious agent shipping in 2026 has solved autonomy and skipped identity, and the workarounds — your card on file, your inbox, your session, your password vault — all put your own identity inside the agent's blast radius. Giving the agent a name of its own, with one balance to fund, one ceiling to set and one switch to pull, is a straightforward answer to that. The plan beyond pre-registration is to open it to OrcaRouter users generally; for now, the name is the thing worth taking.
Sourcing note: OrcaID capabilities, status labels, handle rules and governance controls are from orcaid.ai, checked 2026-08-22, and are the company's own claims — the "By verification" and "At launch" labels are quoted from that page. Grok Bot's launch date, cloud-computer model and sign-in behaviour are SpaceXAI's own description (11 August 2026); the criticism about logging in as you and undisclosed authentication is from independent analyst coverage, not SpaceXAI. OpenClaw's ~355,000 stars, 135,000+ exposed instances, CVE-2026-25253 (CVSS 8.8) and supply-chain campaign are third-party reported. Hermes Agent's model-agnostic design and the Bitwarden/1Password integration in v2026.8.16 are from Nous Research's own release notes. The wider OrcaRouter rollout beyond pre-registration is a stated plan, not a published date. No integration between OrcaID and any agent named here has been announced.
