Adds DeepSeek and Qwen presets to both endpoint protocols: OpenAI-compatible (api.deepseek.com, dashscope-intl compatible-mode/v1) and Anthropic-compatible (api.deepseek.com/anthropic, dashscope-intl apps/anthropic).
Co-authored-by: Claude Code <noreply@anthropic.com>
* claim: #8220
* Agent chat: Custom instructions in Preferences, added to the system note (#8220)
Preferences > AI gets a multi-line Custom instructions field, saved with
the other AI preferences (agent_instructions, kept by page resets like
other user-written text). The chat adds the text after its own system
note when a session starts, by the routes #9811 set up: ACP session/new
(the Claude adapter's _meta key, the standard systemPrompt field, the
Antigravity key) and the system message or field of every API protocol.
Connect and New chat set it, so an edit applies from the next chat and
never changes the one running. Blank text adds nothing; at most 4000
characters are kept, in the field, over ui.agent_chat and from a hand-
edited settings file. ui.agent_chat takes and reports `instructions`.
* Custom instructions: typing in the field stays in the field
The text area did not report itself to Preferences search as a key sink, so with
the field focused, search took the typed characters and moved focus away. The
field now notes the sink while it has focus, like the other preference fields,
and the #8220 test types into it.
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
* claim: #10854
* Agent chat: tell the agent the design changed by hand since its last reply (#10854)
The chat remembers the design as the agent last saw it, at the end of a
turn it answered or worked in. When the next message goes and the design
is no longer that one, a note goes before the message for the agent (not
into the conversation, like the note after an interruption): the user
changed the design by hand, which parameters changed from what to what,
which features are new, edited or removed, where the History marker went,
and to read the design again before reusing positions, picks or sizes.
A new chat and a stopped backend forget it, so a session that has seen
nothing gets no note. Each list names at most 12 items and counts the
rest; names and expressions are cut at 60 characters.
* Agent chat docs: the design-changed note asks the agent to read the design again (#10854)
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
* claim: #11527
* Agent chat: a text selection dragged past the edge scrolls the conversation (#11527)
Holding a text selection near or past the top or bottom of the conversation
scrolls it, faster the further out, so a long answer can be selected in one
drag. egui scrolls a label's selection only for the label under the cursor, and
not at all for a table, whose label sits in its own horizontal scroll area.
egui also leaves a label out of the text selection while it is scrolled out of
view: a selection whose start had scrolled away was dropped, and Copy missed
the hidden text. The conversation's labels now take part out of view too,
clipped to nothing, so they keep their end of the selection and give their
text to Copy without being drawn or moving the selection.
* Agent chat: only a drag towards the edge scrolls the conversation (#11527)
A press on a label creates an empty selection on the press frame, so a click,
a double click or a held press on the line by the edge scrolled the
conversation, and so did the start of a drag from there inwards. Scrolling now
waits until the pointer has moved more than 8 points towards that edge since
the press.
* Agent chat: approval titles and pending messages keep a selection scrolled away
They were plain selectable labels, so a selection that started in one was dropped
once its label scrolled out of view, as in the rest of #11527. They now use the
same off-screen-aware label as the conversation's text.
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
A permission question counted as one of SolidCraft's own tools when its title merely contained
"solidcraft". Now the whole title must be a tool of the chat's own MCP server, named as
`mcp__solidcraft-app__<tool>`, and its kind must be `other` or not given. Anything else goes through
the user's mode and remembered answers, or is asked in the panel.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Selectable transcript labels do not request TextEdit focus, so the global
clipboard handler also copied, cut and pasted selected CAD geometry while
the user worked with chat text.
Keep keyboard ownership with the visible chat after a pointer press until
the next press outside it. Release it when the panel is hidden, and do not
use selections from other native viewports as a global CAD shortcut guard.
Cover real label selection, clipboard events, fallback key events, quick
select-all, hidden panels and deliberate return to the CAD viewport.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* ui_kit: search_picker, a filtered dropdown with a custom entry
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: endpoints list their models with names, say when listing fails, and OpenClaw and OpenRouter presets (#2582)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: a typed model goes to endpoints and to agents that offer a model option
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: the header's model pickers search and take a custom model
The popup can't grow once it is open, so search_picker now fixes the height of what changes below
its box when it opens: the custom row or a late list could otherwise end up out of reach.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* AI preferences: pick an endpoint's model when adding it and on its card (#2582)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: model picker rows read from the left, and a refused typed model says which
From running it: list_picker centres its text, which reads badly for a long list of ids, and an
agent's own refusal can say no more than "Internal error". The search box also asks for a repaint
while it waits for the focus, so an idle window does not stop short of it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: docs for the model picker
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: an endpoint's answer that is no model list says so, an empty list too, and editing lists with the kept key
A 200 without a `data` array (a web page where the endpoint should be, as OpenClaw's gateway
serves until its chat endpoint is enabled) was cached as a successful, empty listing, so the
picker stayed blank with nothing to say and never asked again. Now it is an error with its own
note, an empty list says so, and both are asked for again the next time.
The Edit dialog lists its endpoint's models when it opens, with the kept key read from the
keychain on the listing's thread, as the card does; the key stays out of the field.
The method-not-found test expects the refusal notice that names the model, as every refused
model gets since this branch.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* Agent chat: models from several providers fold under their providers inside the picker
As Chemiseblanc does in #10187: with two or more groups, `search_picker` draws each group as a
header with a chevron, the name and how many of its choices are listed, and a click folds or opens
it. Before anything is typed only the current value's group (and the one the caller asks for) is
open; a group with a match opens once something is typed; a fold made by hand stays until the popup
closes. Enter takes the first row shown, so never one folded away. A lone group keeps its plain
heading.
The header's model picker feeds its providers in as groups, so the separate provider drop-down
goes; the control channel's `provider` now says which group opens first. A header click gives the
focus back to the search box, so typing goes on filtering.
The dialog's listing test checks what kind of listing it got rather than translated words, which
the listing thread may render in another language than the test's.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* Agent chat: rustfmt the provider groups
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* Agent chat: an endpoint's models named `Vendor: Model` fold under their vendors
OpenRouter lists `openai/gpt-6.1-terra` as "OpenAI: GPT-6.1 Terra". A listing whose ids read
`vendor/model` and whose names read `Vendor: Model` now names its providers, so the picker folds
it under its vendors like a multi-provider agent; the rows read the model's name, and the id stays
in the tooltip and in the search. An agent's own `Provider: …` word still comes first. One vendor
is one flat list, as before.
The Add dialog's picker has no current model to prefer, so with nothing to prefer every group
starts open; folding is a click away.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #9619
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Settings: edit an added AI agent (#9619)
Agents you added in Preferences > AI get an Edit button between Test and
Remove. It reopens the Add agent dialog filled with the agent's name and
command, or its protocol, preset, endpoint, model and request timeout, and
Save changes the entry in place: same id, permission override and
always-allowed tools. The stored key is never read back into the dialog;
an empty key field keeps it (still only sent to HTTPS or this computer),
a new key replaces it. The control channel gets `edit_dialog`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #9293
* Assistant: a sign-in confirmed in one tab holds in every tab (#9293)
The latest sign-in evidence per agent is window state: keep it across
tab switches, new tabs and replaced designs instead of resetting it with
the tab's conversation, so a new tab no longer marks the agent as signed
out when discovery's login probe is older than the chat's proof.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #9034
* Assistant: say when an installed agent can't start, and how to reinstall it (#9034)
An agent whose program is found but fails its `--version` check without
printing a version (npm's opencode.exe left as a placeholder when the
package's setup script did not run, which Windows refuses with "not
compatible with the version of Windows you're running") was listed like
a working one, and connecting only said "The agent stopped." with the
system's message.
Discovery now keeps why the program did not start: the spawn error
(Windows' `%1` replaced with the program) or, through a `.cmd` shim, the
first stderr line or the exit status. A probe that times out still says
nothing. The picker shows "Installed, but it does not start: ...", and
connecting first explains it with the agent's install command and Refresh.
The control channel status carries `problem`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #3305
* fix(agent-chat): support ACP configuration options (#3305)
Handle current select options for model, thinking effort and mode, with confirmed snapshots and legacy compatibility. Full quality gate attempted; draft delivery authorized with unrelated geometry/property/timing failures documented in PR #3867.
* fix(agent-chat): keep configuration selectors inline (#3305)
Use compact value pickers and bounded wrapping to preserve conversation space; verified at normal and narrow widths with the installed OpenCode V2 server.
* style(agent-chat): format tool events
* claim: #9495
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Assistant: say when a reply hit its length limit or was declined, not that the model lacks tools (#9495)
The API client never read why an answer stopped, so a Claude reply cut
off at max_tokens while thinking (its thinking text is empty by default),
or a refusal, ended the turn as normal with nothing said. The chat then
blamed tool support: "The model didn't use SolidCraft's tools".
Reply now carries the stop reason: Anthropic stop_reason max_tokens and
refusal, or OpenAI finish_reason length and content_filter and a message
refusal. Any other value is a normal end. The turn ends with that reason,
which already keeps the tool-calling notice away, and the chat shows
what happened in place of the raw "(max_tokens)", for ACP agents too. A
tool call cut off at the length limit has incomplete input, so it is
neither run nor kept in the history.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* agent chat: OpenCode Go preset with per-model endpoints
One preset (https://opencode.ai/zen/go/v1) covers every Go model: each
model is routed to its endpoint family (chat completions, responses,
messages, Gemini), always with the x-opencode-session header Go
requires. Unknown models fall through protocols when Go names the
mismatch; the conversation and tools translate per protocol.
* agent chat: keep Go reasoning metadata, route all listed models (#5656)
* fix agent chat Go endpoints to documented inference routes
* style: collapse long assert in Go routing test (cargo fmt)
* refactor: share Go tool/args helpers, table-driven protocol routing (no behavior change)
* refactor: share GoCallAccum across Go reply parsers (no behavior change)
* refactor: share part/call extractors across Go history converters (no behavior change)
* refactor: share post_to envelope between post and go_post (no behavior change)
* refactor: pass system note into go_responses_input, drop history clone (no behavior change)
* refactor: lazy Gemini name map, shared go_tools, named tool-call cap (no behavior change)
* refactor: share shaped-tool wrapper across Go tool converters (no behavior change)
* fix: set request_timeout_s in Go test setups for rebased main (no behavior change)
* fix: expect ToolsAvailable first in tool-loop tests (matches new event order)
* simplify: reuse tool_arguments, drop restater comments (no behavior change)
---------
Co-authored-by: Julius <julius@localhost>
Co-authored-by: Jom_g <jom_g@users.noreply.github.com>
Co-authored-by: HoloPort <holoport@localhost>
The app id moves from ai.storyteller.solidcraft to io.tangencylabs.solidcraft: macOS bundle id
(the CLI signs as io.tangencylabs.solidcraft.cli), Linux desktop/metainfo/MIME file names, icon
name, Wayland app_id / X11 class, Flatpak, Snap common-id, nfpm/FreeBSD/Nix/AppImage staging,
the web app's icon paths and the docs.
Existing installs keep working: the Flatpak repository marks both old ids
(ai.storyteller.solidcraft and ai.storyteller.solvecraft) end-of-life with a rebase straight to
the new id, re-marking the solvecraft ref that pointed at the 0.8 id; the manifest grants
read-only access to both old ~/.var/app folders and the first launch copies settings and
autosaves from the newest one; AppStream <replaces>/<provides> and X-Flatpak-RenamedFrom list
both; the AppImage removes menu entries it wrote under either old id. The data and settings
folders (solidcraft / SolidCraft) are unchanged. The MSI UpgradeCode and ProgIDs are unchanged.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7009
* Assistant: connect the last-used agent and open the panel at start (#7009)
Two Settings > AI options, off by default: when SolidCraft starts, connect
the agent started last (once the installed agents are known; a missing one
is not swapped for another) and open the Assistant panel. Not for runs
started with --control.
A connection that never gets ready (no agent, start error, the agent ends
while starting) is a new Failed state: the header reads Connection failed
and Connect reads Retry. The control channel reports it as "failed" and
takes connect_at_start / open_at_start.
* Assistant: keep the start-up agent when a design replaces the blank one (#7009)
Opening a design over the blank design SolidCraft starts with reset its
chat and stopped the agent connected at start. A chat not asked anything
yet now starts over in place and keeps its agent. Turning AI features
off also drops a start-up connection still waiting for discovery.
* i18n: Arabic, Hebrew and Vietnamese for the start-up options (#7009)
ar, he and vi are complete catalogs on main now, so every new English
message needs them.
---------
Co-authored-by: MateuszSychXpl <151755914+MateuszSychXpl@users.noreply.github.com>
* claim: #4846
* Assistant: say so when a turn ends without using SolidCraft's tools
When tools were offered and a completed turn made no tool call and no
change, the chat panel shows a short notice pointing to tool-capable
models: always for an empty reply, and for a text-only reply to an
explicit edit request. Plain answers to questions stay quiet. The
control channel exposes it as turnWarning.
* Fix the Anthropic loop test for the tools event; add ja and zh for the no-tool-call notice
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #4846
* Warn when Ollama models lack tool support (#4846)
Check Ollama model capabilities in the background and cache results per
endpoint and model. Show an advisory warning for models without tools,
expose it through ui.agent_chat, and document the behavior.
Cover capability parsing, warning selection, background probing, caching,
and selection changes with synthetic regression tests.
* Keep the Ollama capability probe out of the web build
The panel never calls it there, so it would only be dead code.
* Wrap the model warning instead of cutting it off
In a narrow panel the end of the sentence, the part that says what to do, was truncated.
* Fix AgentConfig default after rebase onto main
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Move the Ollama no-tools warning into a pending locale file with ja and zh
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
When an agent's models come from several providers (Hermes and OpenCode
name each one "Provider · model" or say "Provider: ..." in its
description), the chat header shows a provider drop-down before the model
drop-down, and the model drop-down lists only that provider's models
without the provider in front. It opens on the provider of the model in
use; picking a provider alone changes nothing. ui.agent_chat takes
provider and reports providers and provider.
* claim: #5439
* Agent chat: each design tab keeps its own conversation (#5439)
* Adapt chat-per-tab changes to the new Pending context and cursor tests
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: no per-message limit by default
The limit is optional now: an Unlimited checkbox on the AI page (checked by default), a
number field when unchecked. A saved 24, the old default, reads as unlimited. The API-key agent's
24-request and the gate's 400-request caps are gone; Stop still cancels.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: add the new messages to the complete Japanese and Chinese catalogs
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #5546
* Agent chat: hand WSL paths to agents started through wsl.exe (#5546)
An ACP agent started through wsl.exe was handed the session's working
folder and the solidcraft-app bridge's command as Windows paths, which it
can't use: the Claude adapter refuses the session, and the bridge never
starts. When the agent's program is wsl or wsl.exe, convert both to their
WSL form (C:\Users\... -> /mnt/c/Users/...) before session/new, in the
chat and in the Test button. The bridge still runs on Windows and reaches
the app at 127.0.0.1.
* Agent chat: a Mammouth AI preset for OpenAI-compatible endpoints (#5815)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: the key test names the Mammouth AI URL too (#5815)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #5009
* Assistant: a request timeout per endpoint for slow models (#5009)
An OpenAI-compatible endpoint now stores how long one model request may
take (request_timeout_s, 300 s by default, clamped to 30 s ..= 1 h). It
is set in the Add agent dialog and in the endpoint's entry under Agents
(a running conversation uses a change from its next request), through
ui.agent_chat add_agent, and reported in the chat's state. A request
that runs out of time says the limit and where to raise it instead of
"timeout: global". The desktop HTTP cap goes from 900 s to an hour.
The tool-call timeout stays 330 s: it covers SolidCraft's own tool work
and the user's Allow, never the model.
* Assistant settings: line up an endpoint's Request timeout with Permissions
The label column of an agent's entry is wide enough for "Request timeout",
so the label isn't cut and both rows share one column.
* claim: #2224
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat: keep Send in view for a long message, open "View attached" (#2224)
The message box sat in a fixed 92-point reserve, so a long message grew it
past the panel and took Send, Stop and Attach view with it. The box is now a
panel of its own at the bottom: it grows upwards up to a cap, then scrolls,
and the conversation takes what is left.
The picture sent with a message is kept on it (once the agent starts on the
message), and the "View attached" caption opens it in a window.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat: a message's pictures as thumbnails with a hover preview and a viewer (#2224)
Instead of a window on the far side of the screen: the pictures of a message
show as small thumbnails under it, hovering one shows it bigger for as long as
the pointer stays, and a click opens a viewer over the app with the message's
pictures one at a time (previous and next, arrow keys, Esc). A picture carries
a note shown with it, so attachments of other kinds can use the same viewer.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* claim: #2286
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat: Attach selection sends what is selected in the model with the message (#2286)
A checkbox beside Attach view. The selection is read when the message is
sent (also for a queued one) and described: faces and edges with the names,
indices and points the tools take plus their size and place, features by name
and kind, bodies by name; at most 64, the rest counted. The agent gets it as
text after the message; the conversation shows the message as typed with the
things as grouped chips (hover for what the agent was told).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* WIP: the chat's context picker dialog (#2286)
* Agent chat: Pick context, a picker for elements, features and bodies (#2286)
A button beside Attach selection opens the picker: a dialog with Elements,
Features and Bodies inputs, so clicks in the model, the timeline and the
browser pick into the active one (clicking again takes one out), with the
usual highlight. OK puts the picks as chips with an x above the message box;
they go with the next message like an attached selection (text for the agent,
chips in the conversation) and the tray is empty again afterwards. Cancel
keeps what was picked before. The control channel opens it with pick.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat: Attach selection counts and describes what was clicked, not the pieces of a rim (#2286)
A click on a circular rim or a split surface selects all its pieces for the
commands. The checkbox counted them one by one and the agent got one entry
per piece (reported on macOS). They are now grouped back into the click they
came from: one chip, one entry with the whole size, and the indices of every
piece in its detail. Edge details are looked up by position as well as
index.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat: a message's picture is kept on exactly that message (#2224)
The turn now carries its message's place in the conversation (also through a
queue and a send again after a sign-in), so its picture is kept there, once,
and never on an older message that also wanted one (review of #2225).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat: use the UI kit's checkbox and button and the font tokens (guard tests)
Main's UI-kit guard tests reject raw egui checkboxes, buttons and font sizes in the
chat panel; Attach selection, Pick context, the chip remove button and the chip fonts
now come from ui_kit and Tokens.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* claim: #4851
* solidcraft-cli bench-assistant: the standard Assistant model test
Runs the eleven v1 tasks of #4851 against a desktop app that is already
running with a control port. Each task starts a new design and a new chat
with All permissions, runs only the prescribed setup, sends the prompt
once and waits for the turn. The result is checked on the design
(document.inspect with measurements): body count, volume within 1%
(0.3% for the fillet), bounding-box size in any orientation, the lowest
corner, the body name, live feature kinds, and changed or unchanged
content; the volume question is checked on the reply. Failures get the
issue's categories from the conversation's items. Prints a table, and
writes the issue's comment template (--markdown) and the raw results
with each conversation (--json).
* claim: #5015
Co-Authored-By: Claude Code <noreply@anthropic.com>
* Assistant: add Anthropic-compatible API agents (#5015)
The Add agent wizard now offers Anthropic-compatible endpoints beside OpenAI-compatible ones. The API backend speaks the Anthropic Messages protocol with the same MCP bridge, tool gate, limits, undo and keychain rules: x-api-key and anthropic-version, a top-level system note, input_schema tools, content-block answers and one user tool_result message per round. Test sends a short message because compatible endpoints (Qwen's) need not list models.
Co-Authored-By: Claude Code <noreply@anthropic.com>
---------
Co-authored-by: Claude Code <noreply@anthropic.com>
* claim: #4765
* Agent chat: explain the Flatpak sandbox when no agent is found (#4765)
In the Flatpak the sandbox has no home-folder or network access, so agents
installed through nvm or ~/.local/bin are never discovered and the chat only
said that no agent was found. With FLATPAK_ID set and nothing discovered, the
picker and Preferences > AI now say why and offer the exact
`flatpak override --user ai.storyteller.solidcraft ...` command with a Copy
button. The message is in every catalog, and docs/agent-chat.md gains a
Flatpak section listing what each permission is for. The manifest is
unchanged.
* Vertical history strip: left of the browser, compact rows, hollow-circle marker, collapsed rail hover
* Vertical history strip: seam handle to expand and collapse names, drag to resize with snap
* Vertical history strip: no empty gap below the transport
* Show the history seam arrow only on hover (#2028)
* claim: #2021
* Redesign the Assistant chat panel with shared UI components
* Show Assistant in the shared floating panel shell
* Extend floating Assistant to viewport height and shift navigation
* Dock Assistant before history and match docking preview bounds
* Align Assistant top margin with floating shell cards
* Extract a shared panel drag grip
* Rebuild responsive history sizing and three-way docking
* Share browser and History collapse hover styling
* Show the History title as soon as its text fits
* Animate stacked browser and History expansion and collapse
* Make panel folding responsive and independent of frame delays
* Let folding panel contents shrink continuously to header height
* Use four explicit History transport actions in every layout
* Resize the shared Browser and History column from either card
* claim: shared compact-mode containers (#2029)
* Centralize layout-aware UI containers and enforce their use
* Keep docked Assistant at the outer edge beside tool sheets
* Use shared surface policy for Assistant tool space
* Let tool sheets drag over shell panels while avoiding Assistant
* Update Parameters drag bounds assertion for shell overlap
* Keep dragged tool sheets below ribbon and workspace selector
* Enforce universal UI containers in mandatory repository checks
* Remove legacy report windows and require shared window composition
* Remove retired UI styling and require shared controls
* claim: compact layout preference (#2029)
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* ui: spacing, radius and type tokens, per-tab accents, agent and on-accent tokens
* ui: per-tab accent behaviour and a larger application bar with a Start page button
* icons: silhouette master and flat assistant and agent link icons (open for review)
* icons: apply review notes (sparkle assistant, refined plug, silhouette halo)
* icons: wire the silhouette master to inspect.silhouette; record the accepted icons
* ui: application bar to the design: separators, Save, house, assistant icon, readable hovers
* ui: ribbon structure: Model to View tabs, contextual Sketch and Sheet Metal, migrated pins
* ui: ribbon painting to the compact design
* ui: smaller icon button highlight and assistant mark on the bubble corner
* ui: short ribbon labels that wrap at spaces
* ui: ribbon buttons fit their labels and Finish Sketch shows its full name
* ui: constraint gallery, workspace menu, section list headings and richer tooltips
* ui: lighter selected tab label, ribbon surface and dimmer bar separators match the dark design
* ui: document tabs as rounded chips with a dimmer inactive label
* ui: document tabs fit their names and show close only when active or hovered
* icons: close and plus masters for the document tabs (open for review)
* ui: centre the new-tab plus and overflow button on the bar
* ui: a quiet rule under the ribbon tabs on non-contextual tabs
* ui: the application bar mark separates its faces with the bar colour, not white
* ui: open ribbon section label uses the selected tint and a chevron
* ui: dim the tab close and plus icons until hovered
* ui: every ribbon dropdown uses the chevron, the workspace card included
* ui: the active document tab uses the design's surface colour
* ui: the bug button's right padding mirrors its icon side
* ui: square constraint cells and a rule before the grid
* ui: constraint cells back to their earlier size, keeping the rule
* ui: smaller constraint cells and icons, centred in the section
* ui: constraint cells fit the height of the large buttons' icon and label
* ui: sketch Create titles share the other labels' line
* ui: format the application bar and ribbon changes
* ui: active tab surface token and the colour and hover tests that follow the redesign
* tests: the merged parameter and narrow toolbar tests follow the compact ribbon
* i18n: the Malay Gap label has an English source and its Japanese and Chinese translations
* apply_off_thread tests: the evaluation lingers in tests, so still computing never depends on machine speed
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* claim: #2592
* fix(agent-chat): the design bridge is named solidcraft-app (#2592)
A user's own MCP entry named `solidcraft` (as docs/mcp.md suggests) took the
bridge's place in agents that keep their own entry on a name clash, so the chat
worked on a headless session instead of the open design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Grok Build is added as one more entry in the agent discovery list, over ACP, like the others. Thanks @cfaulkingham.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* Agent chat prototype: drive an ACP agent the user already has from a side panel
A desktop-only chat panel (off by default) that starts a coding agent over the
Agent Client Protocol and gives it SolidCraft's MCP server, bridged to the open
window through the token-protected control channel with the default folder
limits. It finds installed agents (Claude Code and Codex adapters, Gemini CLI,
Copilot, Qwen, OpenCode, goose) with version and login state, streams text,
tool calls and plans, shows the agent's permission requests, and when an agent
isn't signed in shows its own login methods without handling any credential.
Design notes and findings: docs/design/in-app-agent-chat.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: say only what the end-to-end run showed
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: gate, one undo per message, docking, process groups, discovery polish
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: Settings → AI, several agents with a switcher, OpenAI-compatible endpoints, permission modes, toolbar button
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: translations, tool cards, Add agent dialog fixes, test agent saves on its own
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: third-party licences for the keychain crates; clippy
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: badge text from the theme tokens
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Agent chat: Auto-allow model edits is the default permission mode
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Jackson Oursland