Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -16,3 +16,7 @@ _tmp/

# agent runtime output (plans the agent writes during /plan runs)
.hermes/

# packed npm tarballs (npm pack output) — the registry is the distribution
# channel, not the repo
*.tgz
14 changes: 13 additions & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,18 @@ Two things make it different from the rest of the terminal-agent shelf:

**Aether Agent is in beta.** Updates are shipping quickly.

> **What `npm i -g` gives you today: 0.1.0.** The npm `latest` dist-tag is still
> `0.1.0`; **0.2.0 is on `main` but not yet published**. So the features described
> under [v0.2.0 in the release notes](RELEASE_NOTES.md) — portable handoffs,
> `--resume` reaching the brain, `aether agent --local "<task>"` working straight
> after an install — are **not** in the package the command above installs. Until
> the 0.2.0 release is cut, build from source to get them:
>
> ```bash
> git clone https://github.com/AetherAI3/aether-agent
> cd aether-agent && npm ci && npm run build && npm link
> ```

## Carry the work across models and machines

```bash
Expand Down Expand Up @@ -89,7 +101,7 @@ aether agent --local # …same terminal, nothing leaves the machi

> Running the separate Python brain instead — Aether's open-source (Apache-2.0) **[Unlimited Context](https://github.com/AetherAI3/Unlimited-Context-LLM)** engine, which gives an Ollama model a billion-token working memory — is opt-in with `AETHER_LOCAL_BRAIN=python` once you have installed it. It is not bundled with the npm package.

> Prefer the installer UI? Download [`install.sh`](install.sh) or [`install.ps1`](install.ps1), inspect it, then run it locally. Set `AETHER_VERSION=0.2.0` (shell) or `-Version 0.2.0` (PowerShell) to pin an exact release. The canonical npm command above verifies registry integrity and disables lifecycle scripts; there are no native or runtime dependencies and no daemon.
> Prefer the installer UI? Download [`install.sh`](install.sh) or [`install.ps1`](install.ps1), inspect it, then run it locally. Both follow npm's `latest` dist-tag, which today is **0.1.0**; `AETHER_VERSION=<version>` (shell) or `-Version <version>` (PowerShell) pins an exact **published** release, so a version that is not on the registry yet — 0.2.0 included — will fail with `No matching version found`. The canonical npm command above verifies registry integrity and disables lifecycle scripts; there are no native or runtime dependencies and no daemon.

## Models & pricing

Expand Down
80 changes: 46 additions & 34 deletions RELEASE_NOTES.md
Original file line number Diff line number Diff line change
@@ -1,3 +1,49 @@
# Aether Agent v0.2.0 — the work outlives the session

**August 19, 2026**

Start a task on one model. Finish it on another, on another machine. Your tests
decide when it's done.

- **Handoffs** — `aether resume export` writes one portable file: the task, the
model that ran it, the verify gate's verdict, how many tests were still
failing, the files that changed, the verification command, and the repository
it belongs to. Copy it anywhere and continue with `aether agent --resume
<file>`, on whatever model you want. Nothing in it is keyed to an absolute
path, so the receiving checkout does not have to live where the work started —
and no file contents, shell commands, or credential-shaped values ride along.
- **`--resume` now reaches the brain** — the prior session is summarized into a
continuation brief the model reads before its own instruction, rather than
being replayed only for the human. With no new task, the run continues the
original one. No re-pasted chat history.
- **`aether agent --local "<task>"` works after a plain npm install** — the
one-shot offline form used to spawn the separately-installed Python brain and
die with `spawn python ENOENT`. It now drives the Ollama brain that ships in
the package. `AETHER_LOCAL_BRAIN=python` opts back in.
- **Session logs stopped redacting your file paths** — the credential filter
matched `pat` inside `path`, so every edited file in every log read
`[REDACTED]`. Real credential keys are still redacted.
- **A run that never reached your tests no longer reports a failing one.**
- **`npm run demo:handoff`** — a five-second deterministic proof of all of it:
two sessions, two models, two checkouts, one verify gate, no account and no
model download. See [`docs/demo/handoff.md`](docs/demo/handoff.md).

**Availability.** 0.2.0 is on `main`, but it is **not yet published to npm** —
the registry’s `latest` dist-tag still resolves to 0.1.0, so a plain
`npm i -g aether-agents` installs 0.1.0 and none of the above. Until the 0.2.0
release is cut, build it from source:

```bash
git clone https://github.com/AetherAI3/aether-agent
cd aether-agent && npm ci && npm run build && npm link
```

Once 0.2.0 is published, `npm i -g aether-agents --ignore-scripts` will upgrade in
place: no configuration changes, no migration, and 0.1.x session logs are read
unchanged.

---

# Aether Agent — your artifacts get an identity, and doctor stops guessing

**August 14, 2026**
Expand Down Expand Up @@ -37,40 +83,6 @@ Nothing to do on upgrade: `output open <number>` keeps working and existing
history migrates on first read. Full detail in
[docs/releases/2026-08-14.md](docs/releases/2026-08-14.md).

---

# Aether Agent v0.2.0 — the work outlives the session

**August 19, 2026**

Start a task on one model. Finish it on another, on another machine. Your tests
decide when it's done.

- **Handoffs** — `aether resume export` writes one portable file: the task, the
model that ran it, the verify gate's verdict, how many tests were still
failing, the files that changed, the verification command, and the repository
it belongs to. Copy it anywhere and continue with `aether agent --resume
<file>`, on whatever model you want. Nothing in it is keyed to an absolute
path, so the receiving checkout does not have to live where the work started —
and no file contents, shell commands, or credential-shaped values ride along.
- **`--resume` now reaches the brain** — the prior session is summarized into a
continuation brief the model reads before its own instruction, rather than
being replayed only for the human. With no new task, the run continues the
original one. No re-pasted chat history.
- **`aether agent --local "<task>"` works after a plain npm install** — the
one-shot offline form used to spawn the separately-installed Python brain and
die with `spawn python ENOENT`. It now drives the Ollama brain that ships in
the package. `AETHER_LOCAL_BRAIN=python` opts back in.
- **Session logs stopped redacting your file paths** — the credential filter
matched `pat` inside `path`, so every edited file in every log read
`[REDACTED]`. Real credential keys are still redacted.
- **A run that never reached your tests no longer reports a failing one.**
- **`npm run demo:handoff`** — a five-second deterministic proof of all of it:
two sessions, two models, two checkouts, one verify gate, no account and no
model download. See [`docs/demo/handoff.md`](docs/demo/handoff.md).

Upgrade with `npm i -g aether-agents --ignore-scripts`. No configuration
changes, no migration; 0.1.x session logs are read unchanged.

---

Expand Down
Binary file removed aether-agents-0.2.0.tgz
Binary file not shown.
Loading