Skip to content

fix(agent): pin SDK background model to stop retired-snapshot 404s #757

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
posthog wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

fix(agent): pin SDK background model to stop retired-snapshot 404s #757

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions src/lib/agent/agent-interface.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@ import {
WIZARD_USER_AGENT,
WIZARD_WARLOCK_DISABLED_FLAG_KEY,
DEFAULT_AGENT_MODEL,
DEFAULT_HAIKU_MODEL,
} from '@lib/constants';
import {
type AdditionalFeature,
Expand Down Expand Up @@ -970,6 +971,19 @@ export async function runAgent(
// Drop any shell ANTHROPIC_API_KEY so it can't override the wizard's
// OAuth gateway token.
ANTHROPIC_API_KEY: undefined,
// Pin the SDK's small/fast background model. The main query() call
// pins the primary model (claude-sonnet-4-6), but the Claude Agent
// SDK uses a separate small/fast model for background work (topic
// detection, summarization, quota probes). Left unset, the SDK falls
// back to its built-in default — a retired claude-3-5-haiku snapshot
// the PostHog LLM gateway doesn't serve — so every background call
// 404s with not_found_error. It's silent: those rows carry no
// $ai_http_status, so it only surfaces in AI observability. Pin to
// gateway-supported bare aliases (both the current env name and the
// legacy one) so the SDK can never reach for a retired snapshot.
ANTHROPIC_DEFAULT_HAIKU_MODEL: DEFAULT_HAIKU_MODEL,
ANTHROPIC_SMALL_FAST_MODEL: DEFAULT_HAIKU_MODEL,
ANTHROPIC_DEFAULT_SONNET_MODEL: DEFAULT_AGENT_MODEL,
// Defer MCP tool schemas to avoid bloating the system prompt.
// The posthog-wizard MCP exposes many query tools with large schemas;
// without deferral these consume ~113k tokens upfront, leaving
Expand Down
9 changes: 9 additions & 0 deletions src/lib/constants.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,15 @@ export const DEFAULT_AGENT_MODEL = 'claude-sonnet-4-6';
*/
export const HAIKU_MODEL = 'claude-haiku-4-5-20251001';

/**
* Background/small-fast model for the Claude Agent SDK's auxiliary operations
* (topic detection, summarization, quota probes). Pinned via env vars so the
* SDK never falls back to its built-in default — a retired Haiku snapshot the
* LLM gateway doesn't serve. Bare alias (no date) so the gateway's strict
* Bedrock lookup matches, same convention as DEFAULT_AGENT_MODEL.
*/
export const DEFAULT_HAIKU_MODEL = 'claude-haiku-4-5';

// ── Integration / CLI ───────────────────────────────────────────────

/**
Expand Down
Loading