Sprint/v1.2.0 by aaddrriann · Pull Request #3 · pango07/flicky

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann · 2026-04-28T03:19:24Z

Summary

Adds a Local provider path in the Mind tab, routing
inference to any Ollama or OpenAI-compatible local endpoint (LM Studio,
vLLM, xAI, etc.)
New OllamaAPI service handles connection testing, model listing,
pull/delete/create, and streaming chat — with URL normalization to fix
external provider path doubling
Full connection manager UI: AddConnectionModal, ConnectionRow,
OllamaSection, and OllamaManageModal with an 11-entry vision model
catalog
Stream runs to EOS with no timeout or token cap; PTT AbortSignal is
the sole cancellation mechanism — prevents silent hangs on CPU-only
inference
Manage modal polished: widened to 680px, catalog and installed model
list both scroll with a consistent thin 5px pill scrollbar

Test plan

Launch dev build; navigate Mind tab → Local — OllamaSection renders
Add Connection → Verify returns green + model count
Save; row appears with toggle, Manage, Configure actions
Trigger PTT → Ollama responds via streaming chat
Add external connection with bearer token → Verify succeeds (no 404)
Manage modal → all three scroll regions use thin pill scrollbar
Quick-pull a catalog model → progress bar updates; model appears in
list
Kill Ollama mid-stream → error surfaces gracefully, no crash
npm run typecheck — zero errors

Sprint alignment: - Add scripts/sprint-start.sh — fetches origin/master and compares commit counts using --count to avoid Windows line-ending issues; exits 1 on divergence so callers can branch on exit code. - Add .claude/skills/sprint-align.md (/sprint-align) — protocol for running the script, interpreting output, and guiding rebase/merge decisions without executing them autonomously. Branch lifecycle skills: - Add .claude/skills/new-branch.md (/new-branch) — enforces correct base branch per type: feature/* off dev, fix/* off master, sprint/* and release/* off master and dev respectively. - Add .claude/skills/merge-flow.md (/merge-flow) — validates PR target, blocks direct-to-master routing for non-hotfix branches, and runs a pre-PR checklist covering alignment, version bump, and changelog. - Add .claude/skills/commit-msg.md (/commit-msg) — drafts structured commit messages: section labels, em-dash rationale, 72-char wrap, no Co-Authored-By, no markdown in body. Settings and gitignore: - Add .gitattributes — forces LF on *.sh and .claude/skills/*.md to prevent CRLF corruption of bash scripts on Windows checkout. - Add .claude/settings.json — registers .claude/skills/ as the skill discovery path for Claude Code. - Ignore run.bat and setup.bat — local developer convenience scripts with no value in version history.

Type system & storage: - Add LocalConnection interface and extend MindProvider to include 'ollama' — routes inference to any OpenAI-compatible local endpoint alongside Anthropic and OpenAI. - Extend FlickySettings with localConnections array; add IPC channels for CRUD, bearer key management, and model pull/delete/create. Ollama API service (new ollama-api.ts): - testConnection tries /api/tags then falls back to /v1/models — covers native Ollama and external OpenAI-compat providers (xAI, LM Studio). - normalizeBase strips trailing /v1 to prevent URL path doubling when callers append /v1/chat/completions. - streamChat carries no timeout and no token cap — model runs to EOS; PTT AbortSignal is the sole cancellation mechanism, preventing silent hangs on slow CPU inference (e.g. qwen3-vl:8b at ~8 t/s). - isVisionModel covers 16 known vision families for screenshot routing. IPC & preload: - Eight new handlers in index.ts for connection CRUD, bearer key ops, and Ollama model pull/delete/create with progress event forwarding. - Preload exposes matching methods on window.flicky including streaming pull-progress, pull-complete, and pull-error event subscriptions. UI — MindTab, OllamaSection, ConnectionRow, AddConnectionModal: - MindTab gains a third "Local" tab; selecting it renders OllamaSection which lists enabled connections with per-row model selector. - AddConnectionModal: External/Local type toggle, URL verify with latency + model count hint, bearer auth, prefix ID, model IDs, tags. - OllamaManageModal: installed model list, 11-entry vision model catalog with quick-pull, manual pull-by-tag, delete, and create. - vision-models.ts catalogs 11 vision-capable models with Ollama library slugs for one-click pull and Browse links. Polish: - Manage modal widened to 680px; catalog wrapped in 300px scrollable container; thin 5px pill scrollbar applied to catalog, installed model list, and modal body — consistent with app scrollbar style. - Redundant vision badge removed from catalog cards; section title already reads "Vision-capable models".

vercel · 2026-04-28T03:19:29Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
flicky	Ready	Preview, Comment	Apr 28, 2026 3:43pm

OllamaManageModal: - onClick={handlePull} → onClick={() => handlePull()} — prevents MouseEvent from being passed as the optional tag parameter, fixing TS2322 that was failing the CI typecheck step. Release: - package.json version bumped 1.0.3 → 1.2.0 to align with sprint/v1.2.0 Ollama local connections feature.

aaddrriann added 2 commits April 27, 2026 14:24