docs: sync README files with latest repo state#139
Open
yyiilluu wants to merge 1 commit into
Open
Conversation
📝 WalkthroughWalkthroughDocumentation updates across server README files clarify embedding provider routing through shared daemon services with fallback, per-interaction content slicing with token limits, and tool-usage prefixing behavior in interaction history formatting. ChangesServer and service documentation updates
🎯 1 (Trivial) | ⏱️ ~3 minutes
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@reflexio/server/README.md`:
- Line 226: The README description is inaccurate: update the bullet for
service_utils.py to state that format_interactions_to_history_string() slices
interaction.content using REFLEXIO_MAX_INTERACTION_CONTENT_TOKENS via
_resolve_max_interaction_content_tokens() and inserts a truncation marker when
content is sliced, and clarify that the “[used tool: ...]” prefix is conditional
(added only when interaction.content is truthy and interaction.tools_used
exists) rather than always prepended; mention that tool-only turns in tests
include non-empty placeholder content so the marker line still appears.
In `@reflexio/server/services/README.md`:
- Around line 45-46: Update the README paragraph to reflect the actual routing
logic: state that LiteLLMClient.get_embedding() chooses between calling
embedding_service_provider.get_service_embeddings() (HTTP/shared daemon) and the
in-process LocalEmbedder based on
embedding_provider_mode()/should_use_embedding_service(); clarify that daemon
selection requires the daemon /health probe to report a matching active_model
and healthy status, and that setting CLAUDE_SMART_USE_LOCAL_EMBEDDING=1 or
REFLEXIO_EMBEDDING_PROVIDER=local_service|internal_service bypasses the
probe—when bypassed the code may raise EmbeddingUnavailableError instead of
falling back to the in-process embedder.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro Plus
Run ID: 2e6207c4-fdb2-404b-aff4-f25c342bb48c
📒 Files selected for processing (2)
reflexio/server/README.mdreflexio/server/services/README.md
| - `operation_state_utils.py`: Centralized `OperationStateManager` for all `_operation_state` table interactions (progress tracking, concurrency locks, extractor/aggregator bookmarks, simple locks) | ||
| - `deduplication_utils.py`: Shared utilities for LLM-based deduplication (used by ProfileDeduplicator and PlaybookConsolidator) | ||
| - `service_utils.py`: Utilities (`construct_messages_from_interactions()`, `format_interactions_to_history_string()` (prepends tool usage info when `tools_used` is present), `extract_json_from_string()`, `log_model_response()` for colored LLM response logging) | ||
| - `service_utils.py`: Utilities (`construct_messages_from_interactions()`, `format_interactions_to_history_string()` (prepends tool usage info and slices long interaction content via `REFLEXIO_MAX_INTERACTION_CONTENT_TOKENS`), `extract_json_from_string()`, `log_model_response()` for colored LLM response logging) |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Verify format_interactions_to_history_string implementation
# Find the function definition
ast-grep --pattern $'def format_interactions_to_history_string($$$) {
$$$
}'
# Search for tool usage prepending logic
rg -nP 'format_interactions_to_history_string|tool.?usage.*prepend|prepend.*tool' --type=py -C5
# Verify REFLEXIO_MAX_INTERACTION_CONTENT_TOKENS usage
rg -nP 'REFLEXIO_MAX_INTERACTION_CONTENT_TOKENS' --type=py -C3
# Check service_utils.py specifically
fd -t f 'service_utils.py' reflexio/server/services --exec cat {}Repository: ReflexioAI/reflexio
Length of output: 50377
Fix the README behavioral wording for tool prefixes and confirm slicing.
format_interactions_to_history_string() slices interaction.content using REFLEXIO_MAX_INTERACTION_CONTENT_TOKENS via _resolve_max_interaction_content_tokens() (and inserts the truncation marker when over budget); this matches the slicing claim.
Tool-usage “prepending” is conditional: the [used tool: ...] prefix is added only when interaction.content is truthy and interaction.tools_used is present. The tests show tool-only turns are expected to carry non-empty placeholder content (e.g., "(tool call)") so the marker line is still emitted—so the “always prepends” phrasing in the description should be qualified.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@reflexio/server/README.md` at line 226, The README description is inaccurate:
update the bullet for service_utils.py to state that
format_interactions_to_history_string() slices interaction.content using
REFLEXIO_MAX_INTERACTION_CONTENT_TOKENS via
_resolve_max_interaction_content_tokens() and inserts a truncation marker when
content is sliced, and clarify that the “[used tool: ...]” prefix is conditional
(added only when interaction.content is truthy and interaction.tools_used
exists) rather than always prepended; mention that tool-only turns in tests
include non-empty placeholder content so the marker line still appears.
Comment on lines
+45
to
+46
| Embedding requests leave the service layer through `llm/litellm_client.py`; local models (`local/minilm-l6-v2`, `local/nomic-embed-v1.5`) are routed by `llm/providers/embedding_service_provider.py` to the shared embedding daemon when reachable, otherwise to the in-process embedder. | ||
|
|
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Verify embedding routing from service layer through LiteLLM client
# Check litellm_client.py for embedding delegation
fd -t f 'litellm_client.py' reflexio/server/llm --exec rg -nP 'get_embedding|embed' {} -C5
# Verify embedding_service_provider.py routing logic
fd -t f 'embedding_service_provider.py' --exec cat {}
# Confirm model name examples are documented or used
rg -nP 'minilm-l6-v2|nomic-embed-v1\.5|local/minilm|local/nomic' --type=py -C2
# Check daemon fallback logic
rg -nP 'daemon.*reachable|embedding.*daemon.*fall|in-process.*embedder' --type=py -C3Repository: ReflexioAI/reflexio
Length of output: 50375
Align README embedding routing paragraph with actual provider mode logic
- The routing described is consistent with the implementation:
LiteLLMClient.get_embedding()decides between the HTTP embedding service (embedding_service_provider.get_service_embeddings()) and the in-process local embedder (LocalEmbedder) based onembedding_provider_mode()/should_use_embedding_service(). - The model examples
local/minilm-l6-v2andlocal/nomic-embed-v1.5are valid (they’re referenced by the provider-mode logic and covered by tests). - The daemon/fallback behavior in the README is accurate for the default “model-driven” routing, but the docs should also mention that daemon selection depends on the daemon health probe’s
/healthresponse (includingactive_modelmatch), and that settingCLAUDE_SMART_USE_LOCAL_EMBEDDING=1orREFLEXIO_EMBEDDING_PROVIDER=local_service/internal_servicebypasses the probe and can raiseEmbeddingUnavailableErrorinstead of falling back to in-process.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@reflexio/server/services/README.md` around lines 45 - 46, Update the README
paragraph to reflect the actual routing logic: state that
LiteLLMClient.get_embedding() chooses between calling
embedding_service_provider.get_service_embeddings() (HTTP/shared daemon) and the
in-process LocalEmbedder based on
embedding_provider_mode()/should_use_embedding_service(); clarify that daemon
selection requires the daemon /health probe to report a matching active_model
and healthy status, and that setting CLAUDE_SMART_USE_LOCAL_EMBEDDING=1 or
REFLEXIO_EMBEDDING_PROVIDER=local_service|internal_service bypasses the
probe—when bypassed the code may raise EmbeddingUnavailableError instead of
falling back to the in-process embedder.
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.
Summary
REFLEXIO_MAX_INTERACTION_CONTENT_TOKENSin the service-layer maps.Repositories/submodules reviewed
yyiilluu/reflexio-enterpriseReflexioAI/reflexioREADME files updated
reflexio/server/README.mdreflexio/server/services/README.mdValidation
git diff --checkNotes/Risks
/root/reflexio-enterprise/how_to_write_readme.mdand are limited to model-facing code-map README files.Summary by CodeRabbit
Documentation
cloud,local_service,internal_service,inprocess, andoff.