Conversation
added 2 commits
September 28, 2026 19:48
…tudio Preserve original sessions in project SQLite and select budgeted evidence for standard Agents without requiring manual retriever setup. Expose input limits and compression thresholds consistently in both Studio creation flows and in generated Agent configuration. Bound optional embedding work, preserve provider credential boundaries, and retain source lookup after restart. Validation on authorized Devbox: SDK contracts 1157 passed / 5 skipped; frontend 1245 passed; production app/widget build and asset references pass; Ruff and pre-commit including gitleaks pass for source and bundled WebUI. Core context module type audit has no diagnostics; broad Pyright still reports upstream and test-double diagnostics. No live model evaluation or customer Runtime changes. Upstream base: 763865d.
zyn080302
force-pushed
the
feat/default-context-compression
branch
from
September 28, 2026 12:11
465b862 to
db69ba3
Compare
added 16 commits
September 28, 2026 20:23
Bound base and eval dependencies to google-adk >=1.34,<2.3 while AgentKit 0.8.x requires OpenTelemetry <=1.37. Keep the full gate for supported targets and add regressions for incompatible releases, matrix alignment, and lock consistency. ADK 2.9.2 remains unsupported.
Drive deadlines independently of host scheduling while preserving real cancellation, durable batch progress and exact missing-chunk recovery. Keep production deadlines and compression behavior unchanged. Validate 500ms scheduler pauses, three rejected fault mutations and the full offline gate on Python 3.10/ADK 1.34 and Python 3.12/ADK 2.2.
Allow Agent(context_compression={"prepare_index": True}) to prepare authorized
sources before query retrieval using separate time and embedding-call budgets.
Reuse completed SQLite batches without consuming the query budget; preserve
source validation, cancellation, no-embedding fallback and Session originals.
Add 14 regression cases to the required context gate and document direct Agent
usage, preparation costs, limits and opt-out behavior.
Give dependency installation a five-minute step limit and retain the existing three-minute Sidecar test limit within a ten-minute job budget. This prevents cold installation from cancelling the tests before their execution budget. Preserve all test commands, coverage requirements and the fail-closed aggregate. Add a workflow regression to the mandatory release gate.
…nking Preserve source-backed evidence selection and SQLite recovery. Enable bounded preparation and ID-only reranking when embedding is configured; reuse the Agent model for extraction with Ark thinking disabled. Add a two-turn Agent/Runner tutorial and regression proving request compression and complete persisted originals. Update developer guides and include teaching changes in the context gate triggers. Validation: pre-commit; Python 3.10/ADK 1.34.0 and Python 3.12/ADK 2.2.0 each 1447 passed, 5 skipped; focused suites each 132 passed.
This branch has not been deployed
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
1 participant
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.
This change enables recoverable context compression by default in the veADK SDK and Studio, reducing input-limit errors caused by long conversations and large tool responses. Model requests retain relevant evidence and source references, while complete session records remain stored in SQLite by default.
Agents can retrieve original content when needed without rerunning business tools. Compression thresholds are configurable, and developers can explicitly disable compression. Bounded retrieval, local fallback, and final request-size checks improve reliability.