Skip to content

YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context

Fix Luna pipe tool routing

Louis Beaumont33d5637dda01
diff --git a/crates/screenpipe-core/src/agents/pi.rs b/crates/screenpipe-core/src/agents/pi.rs
index 748960032..eb4d8f8db 100644
--- a/crates/screenpipe-core/src/agents/pi.rs
+++ b/crates/screenpipe-core/src/agents/pi.rs
@@ -115,8 +115,20 @@ async fn fetch_models_from_gateway(
let body: serde_json::Value = resp.json().await.ok()?;
let data = body.get("data")?.as_array()?;
- let models: Vec<serde_json::Value> = data
+ let models = gateway_models_to_pi_models(data);
+
+ info!("fetched {} models from gateway", models.len());
+ Some(json!(models))
+}
+
+/// Turn the gateway catalog into Pi's provider catalog. The gateway retains
+/// locked models for UI upgrade prompts, while Pi treats every listed model as
+/// selectable. Omit locked entries here so a pipe never appears to select a
+/// model only for the gateway to silently rewrite it to `auto`.
+fn gateway_models_to_pi_models(data: &[serde_json::Value]) -> Vec<serde_json::Value> {
+ data
.iter()
+ .filter(|m| !m.get("locked").and_then(|v| v.as_bool()).unwrap_or(false))
.map(|m| {
let id = m.get("id").and_then(|v| v.as_str()).unwrap_or("");
let name = m.get("name").and_then(|v| v.as_str()).unwrap_or(id);
@@ -140,10 +152,7 @@ async fn fetch_models_from_gateway(
"maxTokens": 32000,
})
})
- .collect();
-
- info!("fetched {} models from gateway", models.len());
- Some(json!(models))
+ .collect()
}
/// Minimal fallback when the gateway is unreachable.
@@ -3392,6 +3401,38 @@ mod tests {
]));
}
+ #[test]
+ fn gateway_catalog_omits_locked_models_from_pi() {
+ let models = gateway_models_to_pi_models(&[
+ json!({
+ "id": "auto",
+ "name": "Auto",
+ "context_window": 128000,
+ "intelligence": "standard",
+ }),
+ json!({
+ "id": "gpt-5.6-terra",
+ "name": "GPT-5.6 Terra",
+ "locked": true,
+ "context_window": 128000,
+ "intelligence": "highest",
+ }),
+ json!({
+ "id": "gpt-5.6-luna",
+ "name": "GPT-5.6 Luna",
+ "locked": null,
+ "context_window": 128000,
+ "intelligence": "high",
+ }),
+ ]);
+
+ let ids: Vec<&str> = models
+ .iter()
+ .filter_map(|model| model.get("id").and_then(|id| id.as_str()))
+ .collect();
+ assert_eq!(ids, vec!["auto", "gpt-5.6-luna"]);
+ }
+
#[test]
fn test_is_rate_limit_error() {
assert!(is_rate_limit_error("HTTP 429 Too Many Requests"));
diff --git a/packages/ai-gateway/src/handlers/chat.ts b/packages/ai-gateway/src/handlers/chat.ts
index bce07dc87..d8ba3b9e7 100644
--- a/packages/ai-gateway/src/handlers/chat.ts
+++ b/packages/ai-gateway/src/handlers/chat.ts
@@ -43,12 +43,14 @@ export const AUTO_WATERFALL_VISION = [
// lacks). Measured (6/19): gpt-5.4 here cost $590/day vs $177 on flex for the SAME
// traffic — 3.3x more, on the small OpenAI pool, for a marginal IQ gain (54 vs ~50-55)
// that pipes don't need; reverted (#4285→). glm-5 + gemini-3-flash are standard-tier
-// fallbacks for when flex is throttled. Keep Luna out of this tool-heavy lane:
-// GPT-5.6 Luna can serve ordinary background chat, but tool-enabled Pi pipe
-// requests currently fall through to GLM. Retrying it here would add latency
-// without actually serving the pipe.
+// fallbacks for when flex is throttled. Luna is a paid, high-volume fallback
+// immediately after the flex primary: it improves resilience without replacing
+// the proven cache-efficient default for every scheduled run. Pi's streaming
+// tool policy is forwarded by the OpenAI adapter, so this lane can use Luna for
+// tool-enabled pipes as intended.
export const AUTO_WATERFALL_BACKGROUND = [
'gemini-3.5-flash', // flex tier — cheapest on the cache-heavy background mix
+ 'gpt-5.6-luna', // paid fallback for complex / tool-enabled pipe work
'glm-5', // free Vertex MaaS fallback, standard tier
'gemini-3-flash', // near-free safety net
];
diff --git a/packages/ai-gateway/src/providers/openai.ts b/packages/ai-gateway/src/providers/openai.ts
index 6af031855..af1df3789 100644
--- a/packages/ai-gateway/src/providers/openai.ts
+++ b/packages/ai-gateway/src/providers/openai.ts
@@ -218,6 +218,11 @@ export class OpenAIProvider implements AIProvider {
stream_options: { include_usage: true },
response_format: this.formatResponseFormat(body.response_format),
tools: body.tools as ChatCompletionCreateParams['tools'],
+ // Keep the streaming path semantically identical to non-streaming
+ // requests. Pi streams every pipe run; dropping tool_choice here meant
+ // a caller's explicit tool policy (notably "required") never reached
+ // GPT-5.6 Luna even though the model supports function calling.
+ tool_choice: body.tool_choice as ChatCompletionCreateParams['tool_choice'],
};
this.applyGenerationOptions(params, body);
diff --git a/packages/ai-gateway/src/test/auto-cost-attribution.unit.test.ts b/packages/ai-gateway/src/test/auto-cost-attribution.unit.test.ts
index caddf35cf..95c68ef68 100644
--- a/packages/ai-gateway/src/test/auto-cost-attribution.unit.test.ts
+++ b/packages/ai-gateway/src/test/auto-cost-attribution.unit.test.ts
@@ -73,10 +73,10 @@ describe('routing chains stay priceable (no silent $0.01 fallback rows)', () =>
expect(getModelCost('gemini-3-flash', 1_000_000, 0)).toBeCloseTo(0.10, 10);
});
- it('places Luna behind the current chat primaries but keeps tool-heavy pipes on proven models', () => {
+ it('places Luna behind the current auto primaries for text, vision, and pipes', () => {
const luna = 'gpt-5.6-luna';
expect(AUTO_WATERFALL.indexOf('glm-5')).toBeLessThan(AUTO_WATERFALL.indexOf(luna));
expect(AUTO_WATERFALL_VISION.indexOf('gemini-3.5-flash')).toBeLessThan(AUTO_WATERFALL_VISION.indexOf(luna));
- expect(AUTO_WATERFALL_BACKGROUND).not.toContain(luna);
+ expect(AUTO_WATERFALL_BACKGROUND.indexOf('gemini-3.5-flash')).toBeLessThan(AUTO_WATERFALL_BACKGROUND.indexOf(luna));
});
});
diff --git a/packages/ai-gateway/src/test/provider-usage.unit.test.ts b/packages/ai-gateway/src/test/provider-usage.unit.test.ts
index 18c877e2a..424cd0bef 100644
--- a/packages/ai-gateway/src/test/provider-usage.unit.test.ts
+++ b/packages/ai-gateway/src/test/provider-usage.unit.test.ts
@@ -78,6 +78,42 @@ describe('OpenAIProvider usage reporting', () => {
expect(text.trim().endsWith('data: [DONE]')).toBe(true);
});
+ it('forwards Pi-shaped tools and required tool_choice on streams', async () => {
+ async function* stream() {
+ yield { choices: [{ delta: { tool_calls: [{ index: 0, id: 'call_1', type: 'function', function: { name: 'add_one', arguments: '{"a":1}' } }] }, finish_reason: 'tool_calls' }] };
+ }
+ const { provider, calls } = makeOpenAIProvider(async () => {
+ const s: any = stream();
+ s.controller = { abort: () => {} };
+ return s;
+ });
+ const tools: any = [{
+ type: 'function',
+ function: {
+ name: 'add_one',
+ description: 'Adds one to a number.',
+ strict: false,
+ parameters: {
+ type: 'object',
+ properties: { a: { type: 'integer' } },
+ required: ['a'],
+ additionalProperties: false,
+ },
+ },
+ }];
+
+ await new Response(await provider.createStreamingCompletion({
+ ...body,
+ model: 'gpt-5.6-luna',
+ stream: true,
+ tools,
+ tool_choice: 'required',
+ })).text();
+
+ expect(calls[0].tools).toEqual(tools);
+ expect(calls[0].tool_choice).toBe('required');
+ });
+
it('drops stream_options and retries when an OpenAI-compatible server rejects it', async () => {
let attempt = 0;
async function* stream() {
diff --git a/packages/ai-gateway/src/utils/auth.ts b/packages/ai-gateway/src/utils/auth.ts
index 360b32104..055e225ce 100644
--- a/packages/ai-gateway/src/utils/auth.ts
+++ b/packages/ai-gateway/src/utils/auth.ts
@@ -12,14 +12,15 @@ import { activeSubscriptionFilter } from './subscription';
* @returns Promise resolving to boolean indicating if token is valid
*/
export async function verifyClerkToken(env: Env, token: string): Promise<{ valid: boolean; userId?: string }> {
- console.log('verifying clerk token', token);
try {
const payload = await verifyToken(token, {
secretKey: env.CLERK_SECRET_KEY,
});
return { valid: payload.sub !== null, userId: payload.sub ?? undefined };
- } catch (error) {
- console.error('clerk verification failed:', error);
+ } catch {
+ // Never log the JWT or upstream verification error verbatim: worker logs
+ // are broadly accessible operational data and may retain request context.
+ console.error('clerk verification failed');
return { valid: false };
}
}
@@ -208,10 +209,10 @@ async function validateSubscriptionWithId(env: Env, token: string): Promise<{ is
const subs = await subsRes.json() as Array<{ id: string }>;
hasSub = subs.length > 0;
} else {
- console.error('Supabase subscription check error:', await subsRes.text());
+ console.error('Supabase subscription check failed', subsRes.status);
}
- } catch (error) {
- console.error('Error in UUID auth path:', error);
+ } catch {
+ console.error('UUID subscription check failed');
}
// Always return resolvedUserId (clerk_id when available, UUID otherwise)
// so the non-subscribed UUID branch upstream also keys on the same id.
@@ -220,7 +221,6 @@ async function validateSubscriptionWithId(env: Env, token: string): Promise<{ is
// Clerk user IDs - resolve to UUID first, then check subscription
if (CLERK_USER_ID_REGEX.test(token)) {
- console.log('Clerk user ID detected, resolving to UUID:', token);
try {
// Resolve clerk_id to Supabase UUID (has_active_cloud_subscription expects uuid)
const userResponse = await fetch(
@@ -255,8 +255,8 @@ async function validateSubscriptionWithId(env: Env, token: string): Promise<{ is
}
}
}
- } catch (error) {
- console.error('Error checking Clerk user subscription:', error);
+ } catch {
+ console.error('Clerk user subscription check failed');
}
// Not subscribed - don't auto-grant, return false so it falls through
return { isValid: false };
@@ -292,7 +292,6 @@ async function validateScreenpipeToken(token: string): Promise<{ isValid: boolea
if (response.ok) {
const data = await response.json() as { success?: boolean; user?: ScreenpipeUserData };
const userData = data.user;
- console.log('Valid screenpipe user token, user:', userData?.email);
return {
isValid: true,
userId: userData?.clerk_id || userData?.id || userData?.email,
@@ -302,8 +301,8 @@ async function validateScreenpipeToken(token: string): Promise<{ isValid: boolea
console.log('Invalid screenpipe user token');
return { isValid: false };
}
- } catch (error) {
- console.error('Error validating screenpipe token:', error);
+ } catch {
+ console.error('screenpipe token validation failed');
return { isValid: false };
}
}