What does it actually know about me?
Your identity is inferred from what you do, never from what you claim. Attunement is how legible you have become: it rises as more of your real work is observed, and it is yours to read at any time.
$ alter status
~maya · Augmented (L3)
attunement 0.72 rising
sources github, obsidian, discord (3 paired)
discovery on · resolvable by field callers
earned $86.20 to date
Pick a question, and see what the CLI answers with.
What comes with your ~handle
None of these cost anything, and there is no paid tier to upgrade into. The last three are not built yet, and each one says so.
Sessions that can see each other
Every session signed in under one ~handle sees what the others are already working on.
Context that survives the session
Hand your context over when you stop, and the next session picks it up with nothing to paste.
Commits that name who wrote them
Authorship binds to the ~handle in your live session, not to a value sitting in a config file.
A vouch the subject agreed to
One ~handle attests a named competence about another, and it lands only if the subject granted it.
A local background service
An optional background program on your machine that makes your ~handle available to your apps. Its signature is checked before it runs.
Pair your own sources
You pair each source yourself, and nothing is read from one you did not pair. Most prove they are yours with a signed challenge or a string you publish, not a login.
Skills for the AI tools you use
The CLI installs skills into the AI tools on your machine. Each skill is published under the ~handle of whoever wrote it.
Listed in MCP directories
~alter is listed in the official MCP registry and 6 other directories, so a client or agent searching them finds it.
Check a folder for damaged files
Run alter doctor loom on a folder and it flags corrupted text, broken links and passwords left in plain text. Nothing leaves your machine.
Knowing which agent is free
You can grant another ~handle permission to see when you are available. Showing that live is not built yet.
Choosing a skill without guessing
Every skill publisher is a ~handle, so you can look up who wrote a skill. Reading what others vouch for them before you install is not built yet.
Which session spent that
Sessions and agents get their own ~handles, separate from people. A record of what each one spent is not built yet.
Install in three commands
Install the CLI, then either register a new ~handle or log in to one you already have. The third command, alter wire, sets up the AI tools on this machine listed below.
# matched to your platform, the other platforms each get their own line
curl -sSf https://truealter.com/install.sh | bash
alter register # or alter login
alter wire- Claude Code
- Cursor
- Claude Desktop
- VS Code
For any other MCP client, paste this URL. ~alter runs as a hosted MCP server.
https://mcp.truealter.com/api/v1/mcp
Call it from your own code
The SDK is the other way in: one typed method per tool, so you write no client and keep no schema in sync. It calls the same MCP server every client above uses.
npm install @truealter/sdk
import { AlterClient } from '@truealter/sdk';
const alter = new AlterClient();
const who = await alter.resolveHandle('~example');The read tools are free to call, and what they return is bounded by what the subject consented to share. The deeper reads carry a price, $0.01 to $1.00 per call, paid in USDC on Base, and 75% of every paid call goes to the person being read. An agent that already has USDC on Base can pay for these calls. The SDK enforces a cap per query and checks the network, the asset and who is being paid. It has no wallet of its own, so you sign with the wallet you already use.
The limits: payments run on Base in USDC only. There is no swap, no bridge and no cash-out to a bank. ~alter records what it owes and to whom, and never holds the money.
Whatever you build, it uses the same calls: one to read an identity and one to pay for the read.
Give your agent its own ~handle
Everything above assumes a person at the keyboard. Software that acts on its own can hold its own ~handle, with no human account behind it. Whatever it talks to can identify it, and it works with other ~handles under its own name, not the name of whoever ran it.
Start at the manifest. It names the endpoint, what costs nothing, and the two calls that create the ~handle. After that it is the same MCP surface the clients above connect to, so there is no second stack to build against.
GET https://truealter.com/.well-known/mcp.json
call register_autonomous_challenge, then register_autonomous · auth: none · price: 0
An agent acting for you is not you
Software acting for you is a different case from an agent with its own ~handle, and the rule for it is decided: it never speaks under your ~handle. It gets its own ~handle in the Bot tier, which ends in .bot. That is a separate name, not yours with a suffix.
You stay the ~handle that is known and that earns. The agent acting for you is visibly not you, and it can do less than you can, so nothing it does reads as something you did.
This is not available yet. No ~handle in the Bot tier can be created today, and a ~handle claimed now cannot become one. The rule applies once it can. The full naming scheme is in the ~handle protocol.
Reference
registries8
v0.5.14, DNS-authed via _alter.truealter.com.
Hosted-connector tier, streamable HTTP transport.
Source-linked server tier, reads the public SDK repo.
Listed under the ~alter account. Connection URL and config are on the Smithery page.
Independent MCP directory. English, 中文, 日本語.
Skill listing for OpenClaw and ClawHub agents.
AI Agent Marketplace listing.
Cursor IDE plugin listing.
source7
TypeScript client. Install from npm as @truealter/sdk. Apache-2.0.
The alter command line. Claim a ~handle and get paid when it is read. Apache-2.0.
Local daemon that keeps your ~handle available to the apps on your machine. Install from PyPI as alter-runtime. Apache-2.0.
brew install true-alter/tap/alter. Apache-2.0.
Runs local Ollama models as an MCP server, so an agent can send work to a model on your machine instead of an API. Apache-2.0.
n8n nodes that resolve a ~handle and verify who someone is inside a workflow. Apache-2.0.
Obsidian plugin that pairs your vault to your ~handle. Your notes stay on your machine.
docs7
Your first call: the MCP handshake, then verify_identity.
How the pieces of the ~alter protocol fit together, and how to call them.
Read tools are open to call. What they return is bounded by what the subject consented to share. Messaging tools need a per-peer grant on top of that.
identity_read and identity_write carry the same consent boundary as the MCP read tools. Scoped to what the subject granted, revocable any time.
Free tools and paid tools, $0.01 to $1.00 per call. USDC on Base. 75% of every paid call routes to the data subject.
Three-tier namespace: Sovereign, Bot, Instrument. Resolution via DNS-TXT discovery.
Every source pairable to a ~handle, grouped by how each proves control. Most need no application registered and no credential held: a signed challenge, or a string you publish yourself.
standards20
- draft-morrison-mcp-dns-discoverydiscovery
Lets an agent find your MCP endpoint by looking up a DNS record.
Defines alter: as the URI for any ~handle.
Pronouns for referring to a person, a collective or an agent in protocol text.
- draft-morrison-identity-attributed-commitsattribution
Binds every git commit to a verified ~handle.
- draft-morrison-reviewed-by-trailerattribution
Binds every code review to a verified ~handle.
How a collective publishes its identity policy in DNS.
Naming convention for MCP tool names.
Decides whether an identity inference runs on a server or on your own device, based on how the signal was obtained.
Binds a paid identity read to the subject's consent and pays the subject a share of the price.
- draft-morrison-substrate-observationcoordination
How agents coordinate by watching shared substrate instead of exchanging messages.
How a language model tags each claim with the source that confirmed it, and marks anything unconfirmed as unverified.
- draft-morrison-identity-accordcoordination
How two collectives grant each other delegated access.
- draft-morrison-agent-channel-fan-outcoordination
How the concurrent agent sessions of one identity exchange short messages.
- draft-morrison-binding-moment-envelopedelegation
A common shape for how an agent brings a consequential decision to the person it acts for.
- draft-morrison-live-reference-resolutionprovenance
How an agent checks each fact against its live source at the moment it uses it, and marks it uncertain instead of acting on a stale copy.
- draft-morrison-ot-command-authoritydelegation
Refuses an agent control command to an industrial system unless it proves who is acting, for whom, with what consent, and with human authorisation when the risk requires it.
- draft-morrison-morning-briefcoordination
A signed, short-lived brief of signals, escalations, and decisions passed from one ~handle to another.
Our submission on identifying AI agents from black-box API access.
Our comment on Questions 5 to 7: the rule should name the properties a trustworthy digital identity method must have, not a list of technologies.
Submission on the new right to erasure: deleting the records a platform holds leaves what it inferred from them in place.
contact
Our handle.
Public posts.
Every public repository.
One of us will read it.