grant/proof-of-work
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v1.1.0:4 — multi-config AI, background generation, ollama auto-detect, system prompt overhaul

Browse Source 
User-feedback-driven release after testing v1.1.0:3. Nine themes:

1. Multi-config persistence
   - New AIConfigProfile table (per-user). Save N configs, toggle one
     active. Switching providers no longer wipes the previous setup.
   - UserPreferences gains activeAIConfigId; legacy single-config
     columns are mirrored from the active profile so existing reads
     keep working without conditional logic.
   - Idempotent boot migration lifts any existing single-config row
     into a default profile.

2. Ollama auto-detect
   - The "Add config" form probes /api/tags on the StartOS internal
     addresses (ollama.startos / ollama.embassy on :11434). If
     reachable: URL pre-fills, model field becomes a dropdown of
     installed models. Fixes the copy-paste UX.

3. Curated model dropdowns for major providers
   - Claude: Opus 4.7, Sonnet 4.6 (1M ctx), Haiku 4.5
   - OpenAI: GPT-5.5, 5.4, 5.4-mini, 5.4-nano
   - Gemini: 3.1-pro-preview, 2.5-pro, 2.5-flash, etc.
   - "Other (type your own)" stays for niche models.
   - Fixes "I tried gemini-3.0-pro and got 404."

4. Background generation
   - lib/ai/generationRunner.ts: detached runner with in-memory
     pub/sub bus. POST /api/ai/generate kicks it off and returns
     immediately. SSE stream attaches by id. The runner survives
     request cancellation; navigating away no longer kills it.
   - New AIGeneration columns: progressText (in-flight stream),
     durationMs (final wall-clock).
   - Generate UI shows a banner explaining background-safety.
   - History detail page polls progress + renders partial JSON
     live for cross-process resume (page refresh, new tab).

5. System prompt overhaul
   - lib/ai/systemPromptBase.ts: structural contract prepended to
     every template. Forces JSON-only output, library-exerciseId
     usage (kills "exerciseId doesn't belong to this user" errors),
     and per-resistance-exercise suggestedWeight (with-history vs
     without-history variants).
   - aiExerciseSchema + ProgramExercise gain suggestedWeight +
     suggestedWeightUnit. Starting a workout from a ProgramDay
     pre-populates SetLog.weight from the suggestion.

6. Test connection improvements
   - Latency in seconds (was ms — confusing for slow Ollama).
   - Stale "✓ Connected" clears on form change.
   - Per-config Test (no need to activate first).
   - Generous maxOutputTokens for thinking models.
   - Gemini surfaces finishReason on empty response (e.g. "blocked
     by safety filter") instead of generic "empty response."
   - Test endpoint accepts a draft body so you can verify before
     saving + before activating.

7. History detail view
   - Click row → full program tree + exact prompts sent. Apply from
     here without re-generating. Pending rows poll for progress.

8. Sidebar sub-navigation
   - AI: Generate / History / Templates
   - Settings: General / Password / Sessions / AI integration /
     Export / Instance (admin) / Danger zone, with anchor scroll.

9. API key UX
   - "Key saved" indicator on saved configs (was confusing to see
     an empty input after a successful save).

Schema migrations (additive, idempotent in entrypoint):
  - AIConfigProfile table created
  - UserPreferences.activeAIConfigId
  - AIGeneration.progressText + durationMs
  - ProgramExercise.suggestedWeight + suggestedWeightUnit

Tests: 16 new (systemPromptBase, modelMenu, generationRunner). 177
total pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Keysat
2026-05-11 08:09:01 -05:00
parent 8f149d35ab
commit 5e291203a5
35 changed files with 3509 additions and 632 deletions
Download Patch File Download Diff File Expand all files Collapse all files
proof-of-work/app/api/ai/generate/route.ts
+61 -180
View File
@@ -1,48 +1,36 @@
import { NextRequest } from 'next/server';
import { NextRequest, NextResponse } from 'next/server';
import { z } from 'zod';
import { getCurrentUser } from '@/lib/auth';
import { prisma } from '@/lib/prisma';
import { getProvider } from '@/lib/ai/providers';
import {
PROGRAM_OUTPUT_SHAPE,
parseAIProgram,
} from '@/lib/ai/programSchema';
import {
buildHistorySummary,
formatHistoryContext,
} from '@/lib/ai/historyContext';
import { buildBaseSystemPrompt } from '@/lib/ai/systemPromptBase';
import { kickoffGeneration } from '@/lib/ai/generationRunner';
/**
* POST /api/ai/generate
*
* Body: { templateId?: string, userInput: string }
* Body: { templateId?: string, userInput: string, includeHistory?: boolean }
*
* Streams the model response as Server-Sent Events:
* event: generation data: {"id":"...generationId..."}
* event: text data: {"delta":"..."}
* event: usage data: {"tokensIn":N,"tokensOut":M}
* event: complete data: {"parsedOk":true|false,"errorMessage":"..."}
* v1.1.0:4: this endpoint now KICKS OFF a background runner and returns
* the new generation id immediately. The caller subscribes to live
* deltas via GET /api/ai/generations/[id]/stream (SSE) or polls via
* GET /api/ai/generations/[id]. Navigating away no longer cancels the
* generation — the runner keeps writing to the row in the background.
*
* Reads the user's AI provider config from UserPreferences. The full
* library of exercises is appended to the system prompt so the model
* picks real exercise IDs.
*
* On error (no provider configured, model error, etc.) emits a single
* `event: error` and closes.
*
* Always writes one AIGeneration row, regardless of success — so the
* History page can show failed attempts too.
* Response:
* 201 { id: "...generationId..." }
* 400 { error: "..." }
*/
const bodySchema = z.object({
templateId: z.string().optional().nullable(),
userInput: z.string().min(1),
/**
* When true, build + append a compact summary of the user's
* recent (90-day) workout history to the system prompt. Lets the
* model design around stagnations, current strength levels, and
* actual training frequency.
*/
includeHistory: z.boolean().optional().default(false),
});
@@ -51,53 +39,34 @@ export const dynamic = 'force-dynamic';
export async function POST(request: NextRequest) {
const user = await getCurrentUser();
if (!user) {
return new Response(JSON.stringify({ error: 'Unauthorized' }), {
status: 401,
headers: { 'content-type': 'application/json' },
});
return NextResponse.json({ error: 'Unauthorized' }, { status: 401 });
}
const body = await request.json().catch(() => ({}));
const parsed = bodySchema.safeParse(body);
if (!parsed.success) {
return new Response(
JSON.stringify({
error: 'Invalid body',
details: parsed.error.errors,
}),
{ status: 400, headers: { 'content-type': 'application/json' } },
return NextResponse.json(
{ error: 'Invalid body', details: parsed.error.errors },
{ status: 400 },
);
}
// Load the user's AI provider config.
const prefs = await prisma.userPreferences.findUnique({
where: { userId: user.id },
});
if (!prefs?.aiProvider || !prefs?.aiModel) {
return new Response(
JSON.stringify({
return NextResponse.json(
{
error:
'AI is not configured. Open Settings → AI integration and pick a provider + model.',
}),
{ status: 400, headers: { 'content-type': 'application/json' } },
);
}
const provider = getProvider(prefs.aiProvider);
if (!provider) {
return new Response(
JSON.stringify({ error: `Unknown provider: ${prefs.aiProvider}` }),
{ status: 400, headers: { 'content-type': 'application/json' } },
},
{ status: 400 },
);
}
// Load the template if provided, else use a no-op default.
// Load the template if provided.
let template:
| {
id: string;
name: string;
systemPrompt: string;
userPromptTemplate: string;
}
| { id: string; name: string; systemPrompt: string; userPromptTemplate: string }
| null = null;
if (parsed.data.templateId) {
const t = await prisma.aIPromptTemplate.findFirst({
@@ -113,23 +82,15 @@ export async function POST(request: NextRequest) {
},
});
if (!t) {
return new Response(
JSON.stringify({ error: 'Template not found.' }),
{ status: 404, headers: { 'content-type': 'application/json' } },
);
return NextResponse.json({ error: 'Template not found.' }, { status: 404 });
}
template = t;
}
// Load the user's exercise library to embed in the system prompt.
// Library for the prompt.
const exercises = await prisma.exercise.findMany({
where: { userId: user.id },
select: {
id: true,
name: true,
type: true,
muscleGroups: true,
},
select: { id: true, name: true, type: true, muscleGroups: true },
});
const libraryJson = JSON.stringify(
exercises.map((e) => ({
@@ -146,138 +107,58 @@ export async function POST(request: NextRequest) {
})),
);
// If requested, build the workout-history summary block.
// History context if requested.
let historyBlock = '';
if (parsed.data.includeHistory) {
const summary = await buildHistorySummary(prisma, user.id);
historyBlock = formatHistoryContext(summary);
}
// Stitch the final system + user prompts.
const baseSystem = template?.systemPrompt ?? DEFAULT_SYSTEM_PROMPT;
const systemPrompt = `${baseSystem}
// v1.1.0:4 base prompt with output contract + weight rules. Stitched
// BEFORE the template's coaching philosophy so output rules win when
// they conflict.
const weightUnit = (prefs.defaultWeightUnit as 'lbs' | 'kg') || 'lbs';
const isLocalModel = prefs.aiProvider === 'ollama';
const basePrompt = buildBaseSystemPrompt({
weightUnit,
hasHistoryContext: parsed.data.includeHistory,
isLocalModel,
});
const templatePrompt = template?.systemPrompt ?? DEFAULT_TEMPLATE_PROMPT;
const systemPrompt = `${basePrompt}
# COACHING PHILOSOPHY (template-specific)
${templatePrompt}
# OUTPUT SHAPE
OUTPUT SHAPE — emit ONLY a JSON object matching this shape (no commentary, no markdown fences):
${PROGRAM_OUTPUT_SHAPE}
LIBRARY — pick exerciseId values from this list when possible. If you need an exercise the user doesn't have, set exerciseId to null and put the proposed name in exerciseName; the user will resolve it during preview.
# LIBRARY (use these exerciseIds; do not invent ids)
${libraryJson}${historyBlock}`;
const userPromptBody =
template?.userPromptTemplate.replace(/{{userInput}}/g, parsed.data.userInput) ??
parsed.data.userInput;
// Persist the pending row up front so the user can see it in
// history even if the stream dies mid-flight.
const generation = await prisma.aIGeneration.create({
data: {
userId: user.id,
templateId: template?.id ?? null,
templateName: template?.name ?? null,
userInput: parsed.data.userInput,
systemPrompt,
userPrompt: userPromptBody,
provider: provider.id,
model: prefs.aiModel,
status: 'pending',
},
const id = await kickoffGeneration({
prisma,
userId: user.id,
templateId: template?.id ?? null,
templateName: template?.name ?? null,
userInput: parsed.data.userInput,
systemPrompt,
userPrompt: userPromptBody,
provider: prefs.aiProvider,
model: prefs.aiModel,
apiKey: prefs.aiApiKey,
baseUrl: prefs.aiBaseUrl,
});
// Stream the model output as SSE.
const encoder = new TextEncoder();
const stream = new ReadableStream<Uint8Array>({
async start(controller) {
const send = (event: string, data: unknown) =>
controller.enqueue(
encoder.encode(`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`),
);
send('generation', { id: generation.id });
let raw = '';
let tokensIn: number | undefined;
let tokensOut: number | undefined;
let providerError: string | null = null;
try {
for await (const chunk of provider.generate({
apiKey: prefs.aiApiKey,
baseUrl: prefs.aiBaseUrl,
model: prefs.aiModel!, // validated non-null at top of POST
systemPrompt,
userPrompt: userPromptBody,
signal: request.signal,
})) {
if (chunk.type === 'text') {
raw += chunk.delta;
send('text', { delta: chunk.delta });
} else if (chunk.type === 'usage') {
tokensIn = chunk.tokensIn;
tokensOut = chunk.tokensOut;
} else if (chunk.type === 'error') {
providerError = chunk.message;
}
}
} catch (e) {
providerError = (e as Error).message;
}
// Parse + validate the assembled response.
let parsedOk = false;
let parseErr: string | null = null;
let parsedJson: string | null = null;
if (!providerError && raw) {
const r = parseAIProgram(raw);
if (r.ok) {
parsedOk = true;
parsedJson = JSON.stringify(r.program);
} else {
parseErr = r.reason;
}
}
// Persist the final state.
const status = providerError
? 'failed'
: parsedOk
? 'completed'
: 'failed';
const errorMessage =
providerError ?? (parsedOk ? null : parseErr ?? 'Empty response');
await prisma.aIGeneration.update({
where: { id: generation.id },
data: {
rawResponse: raw || null,
parsedProgram: parsedJson,
tokensIn: tokensIn ?? null,
tokensOut: tokensOut ?? null,
status,
errorMessage,
},
});
send('usage', { tokensIn, tokensOut });
send('complete', { parsedOk, errorMessage });
controller.close();
},
});
return new Response(stream, {
status: 200,
headers: {
'content-type': 'text/event-stream',
'cache-control': 'no-store',
'x-accel-buffering': 'no', // disable nginx buffering if proxied
},
});
return NextResponse.json({ id }, { status: 201 });
}
const DEFAULT_SYSTEM_PROMPT = `You are a strength and conditioning coach. The user will describe what they want; you produce a complete training program as JSON.
Constraints:
- Pick exercises from the LIBRARY below by their id. Prefer compound lifts for primary slots and accessories for the back half of each session.
- Keep volume reasonable: 4-7 exercises per session, 60-75 minutes total.
- Use rep ranges that match the goal: hypertrophy 6-12, strength 3-6, power 1-5.
- For each exercise specify sets + reps (range or single) + rest in seconds. RPE is optional but useful for intensity-based programs.
- If the user asks for something a single library exercise can't satisfy, pick the closest fit and add a coaching note explaining the variation.
If you cannot produce a complete program for any reason, emit a JSON object with the durationWeeks and weeks arrays best-effort and add a top-level "description" explaining the gap.`;
const DEFAULT_TEMPLATE_PROMPT = `You are a strength and conditioning coach. The user will describe what they want; design a program that matches their goal, experience, equipment, and time budget. Pick exercises from the LIBRARY and stay close to evidence-based programming for the requested goal (hypertrophy / strength / power / conditioning / general fitness).`;