genMetadata augment PoC (local llama.cpp) - #1215
Draft
eschultink wants to merge 76 commits into
Draft
Conversation
…ents/Augment.java Co-authored-by: aperez-worklytics <75276364+aperez-worklytics@users.noreply.github.com>
Add constructor injection so Dagger can provide AugmentProcessor, and suppress SLF4J/JUL warnings during gateway-core tests. Co-authored-by: Cursor <cursoragent@cursor.com>
Ignore computed functionName during Jackson round-trips and align sanitized fixtures with augment sibling output. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Wire lazy model resolution through instance and shared remote resource paths at runtime, address review feedback on model loading and docs, and add provider tests. Co-authored-by: Cursor <cursoragent@cursor.com>
Move ResourceService to gateway-core, drop redundant NLP wiring helpers, and return SentenceMetadataResult instead of ad-hoc maps. Co-authored-by: Cursor <cursoragent@cursor.com>
…Maven lifecycle. Move OpenNLP download to generate-resources so models are on the test classpath, and replace skipped assumptions with explicit assertions. Co-authored-by: Cursor <cursoragent@cursor.com>
Introduces genMetadata augment processing, Dagger-wired config/backend, non-fatal warning headers, Terraform enable_gen_metadata (4096MB floor + remote resources), and MS Copilot PoC rules/docs. Co-authored-by: Cursor <cursoragent@cursor.com>
eschultink
changed the title
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
| @NoArgsConstructor | ||
| @AllArgsConstructor | ||
| @Getter | ||
| @EqualsAndHashCode(callSuper = true) |
An error occurred while trying to automatically change base from
rc-v0.6.8
to
main
July 24, 2026 18:50
Merge s225-gen-metadata-poc while preserving rc host-module features (external ALB, global enable_remote_resources) and adding per-connector genMetadata wiring. Co-authored-by: Cursor <cursoragent@cursor.com>
Keep rc-v0.6.9 dev-example behavior alongside per-connector remote resource defaults from genMetadata wiring. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…ve conflicts from main
Co-authored-by: Cursor <cursoragent@cursor.com>
…budget. Extend Bedrock/Vertex env, IAM, and budget wiring beyond API connectors so custom_bulk_connectors can opt in, and replace GCP's split budget vars with a single llm_budget object. Co-authored-by: Cursor <cursoragent@cursor.com>
… from metadata. Drop PSOXY_ branding and stop Terraform from injecting GOOGLE_CLOUD_PROJECT/REGION; resolve project via ADC and region via Cloud Run metadata. Co-authored-by: Cursor <cursoragent@cursor.com>
Add METADATA_GEN_MODEL_REGION so model location is configurable independently of the Cloud Function region (required for global/multi-region-only models). Co-authored-by: Cursor <cursoragent@cursor.com>
…ing. Gemini 3.5 Flash defaults to MEDIUM thinking, which shared the 256-token output budget and blew the 15s timeout; switch to google-genai with thinkingLevel=MINIMAL, raise default max tokens to 1024, and align classify prompts/parse recovery to JSON. Co-authored-by: Cursor <cursoragent@cursor.com>
Exclude transcript from the broad meetings pathRegex so the genMetadata rules apply, and guard llm_budget / artifacts_bucket_name access so terraform test works when those are null. Co-authored-by: Cursor <cursoragent@cursor.com>
Prefer the Lite Flash model for classify throughput; full Flash remains available via METADATA_GEN_MODEL. Co-authored-by: Cursor <cursoragent@cursor.com>
Avoid Map.toString() for genMetadata/textDigest outputs so tabular columns contain valid JSON objects. Co-authored-by: Cursor <cursoragent@cursor.com>
Accumulate provider input/output tokens per file, log once at end of processing, and store totals in psoxy-* object metadata (with a post-write GCS metadata patch). Co-authored-by: Cursor <cursoragent@cursor.com>
When a jsonPath matches an object ($ / $[*]), send the Map as LLM corpus and attach +self:genMetadata on it; serialize title/body first so truncation keeps the classification text. Co-authored-by: Cursor <cursoragent@cursor.com>
Default to minimal; pass per-augment level through to Vertex (Bedrock ignores) and cache ChatModels by modelId+thinkingLevel. Co-authored-by: Cursor <cursoragent@cursor.com>
…EVEL. Gemini thinkingLevel is not equivalent to Bedrock Claude extended thinking, so keep it as GCP deployment config (default minimal) instead of a per-augment field. Co-authored-by: Cursor <cursoragent@cursor.com>
Use us.amazon.nova-2-lite-v1:0, surface Bedrock TODOs from aws-host, and add Budgets provisioner perms in psoxy-constants for cost-cap deployments. Co-authored-by: Cursor <cursoragent@cursor.com>
amazon.nova-* foundation-model ids cannot be invoked on-demand; map them to us.amazon.nova-* so genMetadata avoids the CRIS 400. Co-authored-by: Cursor <cursoragent@cursor.com>
Nova rejects Converse outputConfig; rely on prompt + parse/schema gate on Bedrock while Vertex keeps constrained JSON. Co-authored-by: Cursor <cursoragent@cursor.com>
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
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.
PoC for LLM-generated metadata alongside sanitized fields:
genMetadataaugment with local GGUF via llama.cpp, building on augments framework andsentenceMetadata(OpenNLP). Latest commit adds genMetadata processing, warning headers, Dagger wiring, and Terraformenable_gen_metadata.Features
+{field}:genMetadatafrom deployment-configured backend (BETA:local/ llama.cpp)enable_gen_metadataon AWS/GCP host modules —ENABLE_GEN_METADATA, 4096 MB memory floor, remote resources for GGUF weightsMS_COPILOT_GEN_METADATA_AUGMENT) and commented YAML example (11 categories, no confidence)X-Psoxy-Warning(augment-gen-unavailable,augment-gen-inference-failed,augment-output-schema-mismatch)Logistics
Change implications
de.kherud:llama(java-llama.cpp JNI) inpsoxy-coreonly; see gen-metadata-augment.mdenable_gen_metadata(host + per-connector); floors memory and enables remote resource paths when setPSOXY_GEN_*,ENABLE_GEN_METADATA{SHARED_RESOURCE_PATH}/llm/{model}.ggufwhen using local backendTest plan
mvn test -pl gateway-core,core -am(local)terraform validateininfra/examples-dev/awsandinfra/examples-dev/gcpenable_gen_metadata = true, upload GGUF, exercise MS Copilot connector with genMetadata rules enabledX-Psoxy-Warningwhen model missing (no 5xx)Made with Cursor