Chat UI adoption + frontier models, managed/BYOK routing, Atlas graph & onboarding fixes#147
Merged
Merged
Conversation
Overlay the per-component CSS from the pre-thesis (v1.1.116) design onto the @synsci/ui components — the reskin behind restoring the centered conversation look. All 24 files are token-driven, so they render against the current theme's color/radius/font tokens rather than transplanting old colors. The .tsx are untouched, so the recent reasoning + subagent-card fixes in session-turn/message-part survive. Files: accordion, button, card, checkbox, dialog, dropdown-menu, hover-card, icon-button, keybind, line-comment, list, markdown, message-part, notebook-cell, popover, radio-group, select, session-review, session-turn, switch, tabs, tag, text-field, toast.
The adopted v1.1.116 component CSS references an older token vocabulary
(--bg-*, --border-default, --color-semantic-*, --text-{critical,success,
warning}-base, --icon-weak, --font-size-xs). Add an additive compat shim
that maps those names onto the current, more granular tokens (--background-*,
--surface-*, --border-weak-base, --icon-critical/info/success/warning-base)
so the reskin renders in the current palette. No current token is redefined,
so the right pane and other screens that depend on them are untouched.
Replace the thesis full-width chat area + thesis Composer with the pre-thesis chat surface: a centered conversation column (md:max-w-200 mx-auto), the 2px between-turns divider, and a gradient prompt dock wrapping <PromptInput> (the legacy composer, revived — all its context providers are still mounted in app.tsx, it was just orphaned). Adds the sticky back-to-parent header with a ← button when the active session has a parentID, so sub-agent sessions are returnable instead of a dead end. Preserves the current data wiring (turnMessages, activeSession, revert banner), center tabs, and right pane; RightPaneTab stays canvas|terminal (no agents tab). Drops the stale "Agents view" comment.
The orphaned PromptInput had drifted to a near-square 4px corner; the pre-thesis design (and the composer the screenshot shows) is a soft 14px radius. Bump the composer shell to rounded-[14px] and the model chip to rounded-[6px]. Logic untouched — the current file keeps its skills-slash entries and openscience-client wiring over the v1.1.116 original.
Replace the thesis "What are we working on?" welcome (serif prompt + agent chips + suggested prompts) with the pre-thesis NewSessionView: project / branch / worktree context, bottom-aligned in the centered column. Agent and model selection stays in the composer, where the old flow put it. Worktree selection is shared between the empty-state view and the composer via a newSessionWorktree signal.
Tighten the model popover toward the old look: rounded-md (was rounded-lg), a defined border-base edge (was border-weak), and shadow-md (was shadow-lg), with slightly reduced footprint. Row content (source badge, variants) is kept — only the container framing changes.
Port the priority-processing toggle to the composer, next to the effort cycler: a normal/fast ghost toggle shown only for gpt-5.5, sent as `fast` on the prompt (backend maps it to OpenAI service_tier: priority). Resets when the model changes. The effort-level cycler was already present.
Overlay the old chat *rendering* components wholesale so messages, the streaming turn, notebook cells, and the model selector render like the pre-thesis UI — not just the CSS. session-turn / message-part / notebook-cell and the dialog-select-model set are the v1.1.116 originals. The composer stays on the adapted PromptInput (openscience client + skills + the new fast toggle); the old tier selector is gone since openscience dropped tiered agents. Reconciled: SessionTurn no longer receives the current-only onRevertMessage/hideTools props, and message-part re-exports ARTIFACT_TOOL so the science tool-renderer still resolves.
The app-wide :focus-visible rule paints a 2px teal outline on every focused button (!important), which overrode button.css's subtle shadow focus and read as a harsh box on the composer's effort/speed controls. Scope it out of the PromptInput and the conversation so the v1.1.116 buttons keep their own soft focus treatment. Focus stays visible (button.css shadow-xs-border-focus); the outline is untouched everywhere else for a11y.
Curated file-extension matches in markdown and the read/edit/write tool cards open the referenced file in the center-pane file view; bash cards reveal the terminal and re-run there. Wired via a data.openFile callback the workspace provides.
…y gradient The show/hide-steps disclosure triangle now rotates with state, has a larger hit area + hover affordance, and the reasoning parts persist after the reply. The sticky sent-bubble was painting itself with --background-stronger (lighter than the page) plus a fade in that tone — recoloured to --background-base so it sits flush with no band.
…odels The composer picker defaults to a curated frontier set (gpt-5.5, claude-sonnet-5, claude-opus-4-8, claude-fable-5, glm-5.2, kimi-k2.7-code, deepseek-v4-pro/flash); everything else stays reachable in Manage Models. A canonicalKey() folds dots<-> dashes and the z-ai/zai/zhipuai alias so a native id and an OpenRouter slug for the same model match. Falls back to show-all when no frontier model is connected.
…; drop fast toggle Model picker rows get provider icons + smaller consistent type; the agent selector renders 'ml' as ML and its value matches the ghost-button font (14px). The gpt-5.5 normal/fast speed toggle is removed from both composers (it billed priority processing).
Removes the per-request fast param + serviceTier:'priority' mapping and the 'fast' field from the prompt/message schemas, so nothing can request priority (more expensive) processing.
…/redo Sessions sidebar gets a live search, double-click inline rename (via a new sync.session.rename), a sentence-case label, and more button padding. The conversation title in the chat header is editable the same way. Re-exposes the orphaned revert/unrevert as first-class /undo and /redo commands.
Pins the graph to the orbit layout and removes the cards/timeline segmented toggle — one graph view, no timeline.
…ct probe The dev-only /api/thesis middleware defaulted to api.syntheticsciences.ai (which doesn't resolve) instead of the app host the backend already uses, 502'ing the graph. Also mirrors the backend's quiet-200 catch-all so an unrouted /project probe stops spamming 404s.
… switch A prior 'openscience login' syncs a managed openrouter credential (thk_ key on the Atlas proxy) plus enabled_providers:[openrouter]. Both survived a switch to byok, so byok kept routing through the wallet and hid the user's own providers. Under an explicit byok toggle we now drop any provider whose effective key is a managed Atlas key, and skip the synced enabled_providers whitelist.
The new-session screen now explains what to do when no model is connected (add a key with 'openscience keys signin', or use managed with 'openscience login'). The BYOK setup dialog surfaces the ChatGPT/Codex OAuth path instead of leaving it CLI-only.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Aayam Bansal (aayambansal)
pushed a commit
that referenced
this pull request
Jul 8, 2026
Chat UI adoption + frontier models, managed/BYOK routing, Atlas graph & onboarding fixes
Aayam Bansal (aayambansal)
pushed a commit
that referenced
this pull request
Jul 8, 2026
Fixes the Format check: 4 files from PR #147 plus notebook-cell + dialog-manage-models which were already unformatted on main.
Aayam Bansal (aayambansal)
pushed a commit
to codeGaus/openscience
that referenced
this pull request
Jul 8, 2026
Fixes the Format check: 4 files from PR synthetic-sciences#147 plus notebook-cell + dialog-manage-models which were already unformatted on main.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adopts the refined v1.1.116 chat surface into OpenScience and layers on a batch of model-catalog, billing, Atlas-graph, and onboarding fixes.
Chat / message UI
Models & composer
claude-opus-4-8↔claude-opus-4.8,z-ai/zai/zhipuai)ml→ ML, consistent control-row fontserviceTierplumbing + schema)Sessions & Atlas graph
/undoand/redoapp.host (was 502'ing) + quiet/projectprobeBilling / providers
thk_providers and skips the syncedenabled_providerswhitelistOnboarding
Testing