Repository navigation
Releases
v1.0.5
igni-core 1.0.5
The release that carries the end-to-end campaign's findings. Every change
below was found by running igni — the real backend, the real igni server,
the real model — and each one has a test that holds it fixed.
What a reader gets
pip install 'igni-core[decisions]==1.0.5'igni_core— the backend module.igni— the command.igni -m "…"runs one task and exits.igni-cloud— the cloud layer, found through theigni.layersand
igni.storageentry points.
[decisions] adds igni's decisions service (Laya, torch), used by
Bypass mode's last check and by host.decide.
What changed
The window tells the truth about the igni server
The account chip says the igni server is not answering, when it is
not. cloud_connected now follows the last real request the session
made to the server: a connection error, a timeout or a 5xx makes it
false, the next success makes it true. There is no polling and no extra
traffic — the answer comes from a request that was going to happen
anyway. A 401 or a 403 is a credentials problem, not a reachability
one, and stays one: an expired token says the token is expired.
A signed-in person is not pushed to a login form because the server
is down. The sign-in wall follows the account, not the connection: the
chip reads "Account: signed in, but the igni server is not answering"
and marks itself with a dot, and the person stays signed in.
A pending approval survives the tab it was asked in
Closing the tab no longer strands a run that waits for an answer.
attach_session with a session id re-sends that session's paused
approval to the window that attaches, so the card is there and can be
answered.
The session list says which sessions wait. A session with a run
waiting for the person is marked as such, and picking it from the
sidebar attaches it — the card comes back with it.
A prompt left before this backend started is offered at once, rather
than being a message nobody ever sees.
The team and the chart show up while they happen
A team's progress is reported live again. The progress of the
sub-agents and team members reaches the window during the run, so the
chat draws the team card while it is happening and not only after the
session is reopened.
A chart streams into the chat as the visualizer draws it — and a
spec the client cannot draw says so ("Could not draw …") instead of
being silently dropped.
A workflow started from the composer can ask. Its lane that needs a
tool approval reaches the person's window, the lanes settle instead of
staying running, and the run card is there when the chat is reopened.
A project's MCP server is the project's
The MCP page labels a server by where its configuration came from, so
a server in a project's .mcp.json can be revoked from the app, and one
kept under the user's own configuration is not confused with it.
A new or edited .mcp.json takes effect without a backend restart,
and a new chat gets the project's MCP tools without waiting for the MCP
page to ask for them first.
The guard catches a script that is fetched and then run
curl -o x.sh && sh x.sh is refused, as is the wget form of it —
the guard that already stopped curl … | sh now stops the download
that comes first.
The rest of the campaign's findings
- A run's output stays in the run's chat: what a run sends after New
chat or a session switch stays with the run it belongs to. - A chat's history shows the views the agent drew in it, after a switch
or a restart. - A click in a view the agent drew says who it reaches — not the agent.
- A session in another folder runs on that folder's own settings.
igni -cresumes from a shared database as the person who signed in.- A preserved worktree's footer says its changes are committed, so the
merge it names brings them over. - A client that connects after a VM credential refusal is told of it.
ignistops while a run waits on an approval, and on a signal at
once.- A pending prompt can be discarded, and a run paused for an approval
keeps its message pending until the resumed run ends. - A scheduled task fires the hooks of the session as it is now; a
sub-agent spawned after/hooks reloadfires the hooks the reload
left; the backend's run fires the hooks a plugin reload left. - A sub-agent's and a team member's tool calls answer to the same safety
lists as the main agent's, and firePreToolUseandPostToolUse. - A worktree-isolated sub-agent's file and shell tools work in its
worktree. - A team member's pause follows the session's policy and never hangs
headless. - The launcher can hand the window token over a pipe
(IGNI_WS_TOKEN_FD), so a window can be started by something that has
no file to hand. - An extension that does not start says how to fix it, and no session
starts it again;igni ext syncstarts what it stores once, so a
failed upgrade keeps the old version. - The first headless run after an install loads the extension's plugin,
and a headless run's prompt carries what its extensions contribute.
Upgrading
pip install --upgrade 'igni-core[decisions]==1.0.5'puts the new
backend in the same environment.- The desktop app installs this release itself, at the version it is
built for (igni-coreand the app share one version number), and
keeps an install made under the old name.
Verifying it
SHA256SUMS lists every file on the release. Each has GitHub build
provenance, a CycloneDX SBOM, and — when the signing key is present —
a cosign signature over the checksums. See
docs/VERIFYING_RELEASES.md.
Install
Python (CLI / library):
pip install 'igni-core[decisions]==1.0.5'
# or
uv pip install 'igni-core[decisions]==1.0.5'
# ``[decisions]`` adds igni's decisions service (Laya, torch) for
# Bypass mode's last check and ``host.decide``; leave it out on a
# backend that points IGNI_DECISIONS_URL at a shared one.Desktop app: released from
ignite-ember/igni-app
under the same version, which installs this backend on first launch.
Verify
SHA256SUMS lists every file on this release. Each one has
GitHub build provenance (gh attestation verify <file> --repo ignite-ember/igni-core; the bundle is attached for offline
checks), the *.cdx.json files are CycloneDX SBOMs, and
SHA256SUMS.sig, when present, is a cosign signature over the
checksums. See docs/VERIFYING_RELEASES.md.
v1.0.4
igni-core 1.0.4
The first release published under igni's own name. igni-core on PyPI
is this repository's build: the igni_core module, the igni command
and the decisions service. The old name still installs it —
pip install ignite-ember resolves to igni-core==1.0.4 through the
alias (packages/ignite-ember/ — metadata, and a dependency on
igni-core at the same version), and the Homebrew formula
ignite-ember/tap/ignite-ember installs igni-core too, so nothing
that worked yesterday breaks.
It follows ignite-ember 1.0.3, the last release of the monolith
under the old name.
What a reader gets
pip install 'igni-core[decisions]==1.0.4'igni_core— the backend module. Theember_codename is gone;
igni_coreis the name every import uses now.igni— the command (ignite-emberis an alias for it, so old
shell aliases and the formula's link keep working). It runs one task
and exits:igni -m "…". There is no terminal chat UI — the chat is
the desktop app and the web client.igni-cloud— the cloud layer (PostgreSQL storage, the producer
and guest binding), in its own package, found through the
igni.layersandigni.storageentry points.
[decisions] adds igni's decisions service (Laya, torch), used by
Bypass mode's last check and by host.decide. Leave it out on a
backend that points IGNI_DECISIONS_URL at a shared one.
What changed
- The split. The backend is published under its own name, and the
cloud layer sits beside it inigni-cloudrather than inside it. - The wire contract is typed. The messages the backend and the app
exchange have one definition, checked from both sides: the app's
client is igni-core's contract, byte for byte. - One module per concern. The backend is split into modules that
each own one thing — settings, auth, the cloud, the tools, the
transport — behind interfaces the tests can substitute. Nothing
reaches into another's internals. igni login. A command, not a script: it opens the portal's
sign-in page in a browser, takes the token back over a local
callback, writes~/.igni/credentials.jsonand hydrates the group
pack.--token-stdindoes it without a browser.- The Windows sandbox is behind one interface. What it allows is
one thing to state, and one thing to test.
Upgrading
pip install --upgrade 'igni-core[decisions]==1.0.4'puts the new
backend in the same environment. The oldember_codepackage stays
where it was installed; nothing imports it any more.- The command is the new one:
igniandignite-emberboth point at
igni_core.__main__:mainafter the upgrade. - The desktop app installs this release itself, at the version it is
built for (igni-coreand the app share one version number), and
keeps an install made under the old name: the cache directory and
the install marker are still read under their old names until the
next write.
Verifying it
SHA256SUMS lists every file on the release. Each has GitHub build
provenance, a CycloneDX SBOM, and — when the signing key is present —
a cosign signature over the checksums. See
docs/VERIFYING_RELEASES.md.
Install
Python (CLI / library):
pip install 'igni-core[decisions]==1.0.4'
# or
uv pip install 'igni-core[decisions]==1.0.4'
# ``[decisions]`` adds igni's decisions service (Laya, torch) for
# Bypass mode's last check and ``host.decide``; leave it out on a
# backend that points IGNI_DECISIONS_URL at a shared one.Desktop app: released from
ignite-ember/igni-app
under the same version, which installs this backend on first launch.
Verify
SHA256SUMS lists every file on this release. Each one has
GitHub build provenance (gh attestation verify <file> --repo ignite-ember/igni-core; the bundle is attached for offline
checks), the *.cdx.json files are CycloneDX SBOMs, and
SHA256SUMS.sig, when present, is a cosign signature over the
checksums. See docs/VERIFYING_RELEASES.md.