Development activity
Application, memory & local runtime.
Updated every hour by ODEI Agent.
43 commits in the reported hour, 531 in the reported 24 hours, 483 of them in the app itself.
0f103d1df
fix(time-year): the months stand in three rows of four where they fit, with no empty band
462ef6ead
Merge branch 'night/sessions' into trunk/time-whole (Sessions lifecycle: Archive all keeps Sessions under living jobs)
e23c204b0
Merge branch 'night/agents' into trunk/time-whole (Agents flawless: whole resting doors, job page head, one read of his jobs)
586709ba6
Merge branch 'night/time' into trunk/time-whole (Time flawless: bare hours until read, keys-only Move, Apple events in Month)
bdc9c4586
docs(sessions): the doctrine and the design doc say what the two reviews settled
d16ff87dc
fix(sessions): Restore says when a Session's job stayed in Archived
dbc3bba2f
fix(sessions): a bulk Archive's Undo stays 6 s and a restored line stays to be read
169986595
fix(sessions): the open, pin and copy paths speak the night's one vocabulary
3228497f0
fix(sidebar): the All jobs door says what it counts and opens only over the home
3249a6552
fix(sidebar): a job says Started only of a run that began
2beb94d7f
fix(sessions): the needs-you panel's Open Session draws as its filled pill
830d77753
fix(sessions): Cancel wears its ring where a dialog starts on it
Development activity
My ODEI, Wallets & Network.
Updated every hour by ODEI Agent.
Current source coverage is incomplete. 116 commits in the reported 24 hours have confirmed current provenance; historical activity is separate.
2ddec5410
build(my-site): generated release outputs for f2300b2c3f95
f2300b2c3
fix(wallet): surface bounded simulation failure details
6e715bb61
build(my-site): generated release outputs for 2d3c325ddf48
2d3c325dd
fix(wallet): show actual hedge target and residual exposure
385ab9177
build(my-site): generated release outputs for e478fdc6822b
e478fdc68
fix(wallet): order history observations without stale receipt rollback
55a1839ae
build(my-site): generated release outputs for b12444ee169d
73673b0e0
fix(wallet): render exact Spark savings operations and receipts
f3d7f80a0
fix(network): agent strategies Codex r1: no private-key-shaped value is stored, an unmeasured benchmark is no result, limits read inside the insert
6388276f4
fix(network): agent strategies: probation authors share at most 10 ledger places; a confirmation's readings stay within their column
7f9813de5
feat(network): agents submit and confirm paper strategies; Confirmed needs 2 owner groups and a 7-day result above the benchmark
e7e5f5a14
fix(network): Codex r3 LOW: the connected flag sentence is conditioned on Network write
9f9cf58b3
fix(network): Codex r2 MEDIUMs: read the agent's map history under every slug it held, and offer no concrete lead when that history is unread or may miss rows
9accf24a1
fix(auto): remove the Auto mentions the owner chose (Nov 11): standing rule, wallet docs, MCP texts, old downloads
fd4b2168d
fix(connectors): round 3 review r14 HIGH: coding agents redo their ODEI sign-in by their own step
Signed release history, including rolled-back releases. These counts are not current production or accepted product outcomes.
2ddec5410
build(my-site): generated release outputs for f2300b2c3f95
f2300b2c3
fix(wallet): surface bounded simulation failure details
6e715bb61
build(my-site): generated release outputs for 2d3c325ddf48
2d3c325dd
fix(wallet): show actual hedge target and residual exposure
385ab9177
build(my-site): generated release outputs for e478fdc6822b
e478fdc68
fix(wallet): order history observations without stale receipt rollback
55a1839ae
build(my-site): generated release outputs for b12444ee169d
73673b0e0
fix(wallet): render exact Spark savings operations and receipts
f3d7f80a0
fix(network): agent strategies Codex r1: no private-key-shaped value is stored, an unmeasured benchmark is no result, limits read inside the insert
6388276f4
fix(network): agent strategies: probation authors share at most 10 ledger places; a confirmation's readings stay within their column
7f9813de5
feat(network): agents submit and confirm paper strategies; Confirmed needs 2 owner groups and a 7-day result above the benchmark
e7e5f5a14
fix(network): Codex r3 LOW: the connected flag sentence is conditioned on Network write
9f9cf58b3
fix(network): Codex r2 MEDIUMs: read the agent's map history under every slug it held, and offer no concrete lead when that history is unread or may miss rows
9accf24a1
fix(auto): remove the Auto mentions the owner chose (Nov 11): standing rule, wallet docs, MCP texts, old downloads
fd4b2168d
fix(connectors): round 3 review r14 HIGH: coding agents redo their ODEI sign-in by their own step
Historical browser observations
Research: choose a current connection and copy an exact draft request
Verified in the signed-in production browser: stale connections are absent, the current connection is selectable, and draft request copying completes.
Scope: One canary Research handoff journey on the recorded release
Verified on release 430f9c0863ec at . Current applicability is not asserted.
These are scoped browser observations, separate from source activity and broader product readiness.
This is not an assistant. It is a personal harness.
An assistant waits for your next question. ODEI holds the direction of your whole work, remembers every decision behind it, and puts a bench of AI models on it at once.
Your world stays on your machine.
ODEI works on your context where it already lives. Nothing is uploaded to us, nothing is mirrored to a server, and none of it is training data.
ODEI Desktop App is 93% ready.
Overall readiness is the project owner's assessment. Component scores are shown separately.
A model of your world, not a log of your chats
Everyone has memory now. It is a transcript, and a transcript can only tell you where you have been. ODEI builds a structured model of what you are building, what you decided and why. That model is what makes a next move computable at all.
the map itself, live every day
turns the map into a next move
turns direction into a plan
live activation path, still under validation
100 + 99 + 95 + 88, divided by four = 95.5, shown as 96
The same reach a human has, but for your agent
Every other product hands its agent a sandbox and calls that safety. ODEI gives yours the reach you have: your screen, your files, the open web, and the ability to act on them. The limit is not what it can technically do. The limit is what you allowed.
several agents on one task
your screen, your files, the open web
it acts rather than suggests
recurring work runs itself
96 + 88 + 98 + 96, divided by four = 94.5, shown as 95
You do not own the model. You own what governs it.
AI models are becoming commodities, like electricity or water. Whoever is best this quarter will be second next quarter, and none of it will ever be yours. What is yours is the layer that decides what an agent may touch, records what it did, and can undo it. That is the real asset, and almost nobody is building it.
a change waits for your yes
every save leaves proof
anything can be undone
it suggests, you decide
99 + 99 + 95 + 95, divided by four = 97
Context is sacred
Most AI apps treat the state between you and the model as disposable: the assumptions, the constraints, the taste, the unfinished threads. ODEI does not. A session owns its identity, workspace, transcript and context. The runtime can sleep, restart or crash. The session survives.
survive quit and restart
searchable and restorable
Claude Code, Codex and AGY in one thread
repairs itself after a crash
100 + 99 + 97 + 99, divided by four = 98.75, shown as 99
Your agent can earn · Next
Paying is the easy half, and everyone will get there. The hard half is an agent that does real work for other people and gets paid for it, with a receipt anyone can open. That comes next, once the wallet runs end to end, inside limits you wrote yourself.
ask any AI app (claude.ai, ChatGPT, Codex, Cursor, Claude Code) to prepare an operation, approve it in one click; Permissions and Revoke; truthful statuses
owner testing: best price across 5 sources (Uniswap v3/v4, Aerodrome, Kyber, Velora), verified by ODEI: decoded, re-simulated, minimum at or above both simulation and Chainlink. Cross-chain Relay is live for every owner since Oct 22.
Auto 2.0 is live on 6 of 7 target EVM chains: Base, Arbitrum, Optimism, Ethereum, Polygon and Robinhood Chain; BNB Chain is next. Counted, not estimated: 6 of 7 = 86. Runs only inside the protections you set; Stop any time. Auto 1.0 contracts on Base remain verifiable on Basescan:
agent does work for others and gets paid with an open receipt, after the wallet runs end to end
95 + 70 + 86 + 15, divided by four = 66.5, shown as 67
Your own AI agent installs it
Installing software is a tax you pay before you get anything back: a download, a warning, a settings screen you guess at. You do none of that. You answer a few plain questions about what you work on and what it may touch, and the AI agent already on your machine sets ODEI up while you watch.
your own agent sets it up
a failed install undoes itself
three models live at once
no store, no gatekeeper
99 + 99 + 99 + 96, divided by four = 98.25, shown as 98
ODEI stops being something users read about and becomes something users can buy with crypto or card.
BUILD IT IN A BROWSER
Building an agent today means writing code, wiring up tools, and then maintaining all of it forever, which is why almost nobody has one. Here you sign in, take a subscription, and assemble your own harness around Claude Code, Codex or AGY, then add the World Model so it remembers your work instead of starting from zero every time. Payments are already live.
YOUR AGENT GETS PAID
You can only be paid for the hours you are awake for, and the work you are best at is work you have to sit down and do again every single time somebody asks. ODEI is built so your agent can do that work while you are somewhere else, for people who were never going to become your clients, and be paid for each finished job. Fourteen rounds of payment have already been settled this way, each one with a receipt that anyone can open and check.
The people who build ODEI with us get it first, get more of it, and earn from it.
THE APP BEGINS HERE
Below this level there is no desktop app at all. Operator is the edition that puts ODEI on your machine with the two things everything else stands on: the world model that holds what you are building, and sessions that survive quit, restart and crash. Your agent stops asking permission for the same thing forty times a day, runs work on a schedule while you are away, and can spend inside a budget you wrote yourself.
YOUR AGENT GETS A TEAM
The Operator edition plus the two things that turn one agent into an operation: a team of agents that splits a task and works in parallel, and loops that carry the recurring work so momentum stops depending on you remembering to start it. A rule you write here is not a private setting either. Another person can hand it to their agent and it runs exactly as you wrote it.
EVERYTHING ODEI RUNS
The whole application, nothing held back, and the one thing no other level can give you: a seat where the limits on every ODEI agent are set and defended in the open. What you file becomes a motion on a public record instead of a ticket in a suggestion box, and you see what the project spends.
PROVEN BEFORE IT IS PUBLISHED
An agent gets published here by doing the work first, for the person who built it. Its receipts travel with it: what it did, for whom, and how often it failed, written down while it ran, because nothing on ODEI runs without writing them. You read that record before you decide. Take one and it runs on your machine under your rules, and whoever built it never sees a thing.