Skip to content

feat(core,cli): finalize context compression with batched routing#4

Open
thetomzohar wants to merge 12 commits intomainfrom
feature/clean-compression-flags
Open

feat(core,cli): finalize context compression with batched routing#4
thetomzohar wants to merge 12 commits intomainfrom
feature/clean-compression-flags

Conversation

@thetomzohar
Copy link
Copy Markdown
Owner

@thetomzohar thetomzohar commented Mar 18, 2026 •
edited
Loading

Summary

Add support for context compression configurations via new CLI flags:
--compress (cloud context routing using flash compressor) and
--local-compress (local context routing using local ollama models). This
improves multi-turn tool performance by selectively compressing large file
payloads in history while retaining accuracy.

Details

  • Added Flags: --compress proxies file classification to Gemini cloud
    (compressor-2.5-flash-lite); --local-compress proxies it to local models
    over HTTP (getLocalContextCompressionModelUrl).
  • Core Integrations: Enhanced LocalContextCompressionService and updated
    GeminiClient to dynamically trigger compression before fetching chat turns
    based on the configured mode.
  • Batched Routing Optimization: Refactored ContextCompressionService to use a 4-pass batched sequence, reducing LLM overhead by grouping multiple file relevance queries into a single call.
  • Removed Debug Logs: Stripped all temporary telemetry instrumentation from
    experimental commits, ensuring clean integration with the trunk.

Related Issues

None.

How to Validate

  1. Run npm run build
  2. Start CLI with --compress: node packages/cli/bin/gemini.js --compress
  3. Load a large file into context via tool usage (e.g. read_file on
    package-lock.json).
  4. Perform subsequent turns and observe that history is successfully compressed
    transparently by checking the local compression_state.json file in your
    Gemini temp directory.

Pre-Merge Checklist

  • Updated relevant documentation and README (if needed)
  • Added/updated tests (if needed)
  • Noted breaking changes (if any)
  • Validated on required platforms/methods:
    • MacOS
      • npm run
      • npx
      • Docker
      • Podman
      • Seatbelt
    • Windows
      • npm run
      • npx
      • Docker
    • Linux
      • npm run
      • npx
      • Docker

@thetomzohar thetomzohar force-pushed the feature/clean-compression-flags branch 3 times, most recently from 1ef9b2f to b70725e Compare March 18, 2026 17:25
@thetomzohar thetomzohar force-pushed the feature/clean-compression-flags branch from b70725e to bad96cd Compare March 18, 2026 17:46
Tom Zohar added 8 commits March 18, 2026 20:16
fix(compression): address code review comments for PR #4
9c453e6 
1. Fix AbortSignal propagation in local/cloud compression to cancel requests properly.

2. Add proper tests and remove duplicated license header.
refactor(compression): rename LocalContextCompressionService to Conte…
45c52c6 
…xtCompressionService

This commit standardizes the service and config names to reflect that it handles both local and cloud modes.
fix(core): improve context compression structural invariant guard and…
fa19f44 
… telemetry handling

This commit corrects a condition where compression breaks the functionCall/functionResponse adjacency invariant by gracefully ignoring edge-cases. It also cleans up temporary telemetry traces to prepare the code for branch PR submission.
style: format workspace with prettier hooks
b8633df
fix(core): address PR review comments for context compression validation
fddf09a 
- Restore chatCommand and logger compression checkpointing state logic
- Reposition setCurrentPrompt block above pre-route compression invocations
- Append pending mock logic hooks into unit test configurations
- Remove dangling settings load statement whitespace in CLI layer
style(core): remove duplicate license header and move import in logge…
1db97aa 
…r.ts
feat(config): add support for setting cloud compression via settings.…
02028e0 
…json
feat(core,cli): finalize context compression with batched routing and…
6eed3f1 
… cloud settings
@thetomzohar thetomzohar changed the title feat(core): add local and cloud context compression variants via flags feat(core,cli): finalize context compression with batched routing Mar 28, 2026
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 1, 2026
@joshualitt
feat(core): Land ContextCompressionService
01f8d6c 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 1, 2026
@joshualitt
feat(core): Land ContextCompressionService
08040ea 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
@joshualitt joshualitt mentioned this pull request Apr 1, 2026
Merged
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
f4c1144 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
d349237 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
8839a86 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
788d705 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
12dfb33 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
2e32ca0 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
9cfdce6 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
c934c47 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
8e16db8 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
dcbefb4 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
joshualitt added a commit to google-gemini/gemini-cli that referenced this pull request Apr 2, 2026
@joshualitt
feat(core): Land ContextCompressionService
30d28fc 
* Upstreams the ContextCompressionService from: thetomzohar#4
* Some minor tidying.
* Refactor: Rename ContextManager to MemoryContextManager
* Settings: Add experimental.generalistProfile to safely default context management hyperparameters
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@thetomzohar