🫱🏼🫲🏽 refactor: Improve Agent Handoffs (#11172)
Some checks are pending
Docker Dev Branch Images Build / build (Dockerfile, lc-dev, node) (push) Waiting to run
Docker Dev Branch Images Build / build (Dockerfile.multi, lc-dev-api, api-build) (push) Waiting to run
Docker Dev Images Build / build (Dockerfile, librechat-dev, node) (push) Waiting to run
Docker Dev Images Build / build (Dockerfile.multi, librechat-dev-api, api-build) (push) Waiting to run
Sync Locize Translations & Create Translation PR / Sync Translation Keys with Locize (push) Waiting to run
Sync Locize Translations & Create Translation PR / Create Translation PR on Version Published (push) Blocked by required conditions

* fix: Tool Resources Dropped between Agent Handoffs

* fix: agent deletion process to remove handoff edges

- Added logic to the `deleteAgent` function to remove references to the deleted agent from other agents' handoff edges.
- Implemented error handling to log any issues encountered during the edge removal process.
- Introduced a new test case to verify that handoff edges are correctly removed when an agent is deleted, ensuring data integrity across agent relationships.

* fix: Improve agent loading process by handling orphaned references

- Added logic to track and log agents that fail to load during initialization, preventing errors from interrupting the process.
- Introduced a Set to store skipped agent IDs and updated edge filtering to exclude these orphaned references, enhancing data integrity in agent relationships.

* chore: Update @librechat/agents to version 3.0.62

* feat: Enhance agent initialization with edge collection and filtering

- Introduced new functions for edge collection and filtering orphaned edges, improving the agent loading process.
- Refactored the `initializeClient` function to utilize breadth-first search (BFS) for discovering connected agents, enabling transitive handoffs.
- Added a new module for edge-related utilities, including deduplication and participant extraction, to streamline edge management.
- Updated the agent configuration handling to ensure proper edge processing and integrity during initialization.

* refactor: primary agent ID selection for multi-agent conversations

- Added a new function `findPrimaryAgentId` to determine the primary agent ID from a set of agent IDs based on suffix rules.
- Updated `createMultiAgentMapper` to filter messages by primary agent for parallel agents and handle handoffs appropriately.
- Enhanced message processing logic to ensure correct inclusion of agent content based on group and agent ID presence.
- Improved documentation to clarify the distinctions between parallel execution and handoff scenarios.

* feat: Implement primary agent ID selection for multi-agent content filtering

* chore: Update @librechat/agents to version 3.0.63 in package.json and package-lock.json

* chore: Update @librechat/agents to version 3.0.64 in package.json and package-lock.json

* chore: Update @librechat/agents to version 3.0.65 in package.json and package-lock.json

* feat: Add optional agent name to run creation for improved identification

* chore: Update @librechat/agents to version 3.0.66 in package.json and package-lock.json

* test: Add unit tests for edge utilities including key generation, participant extraction, and orphaned edge filtering

- Implemented tests for `getEdgeKey`, `getEdgeParticipants`, `filterOrphanedEdges`, and `createEdgeCollector` functions.
- Ensured comprehensive coverage for various edge cases, including handling of arrays and default values.
- Verified correct behavior of edge filtering based on skipped agents and deduplication of edges.
This commit is contained in:
Danny Avila 2026-01-01 16:02:51 -05:00 committed by GitHub
parent d3b5020dd9
commit 791dab8f20
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 521 additions and 71 deletions

View file

@ -150,7 +150,13 @@ export async function initializeAgent(
const provider = agent.provider;
agent.endpoint = provider;
if (isInitialAgent && conversationId != null && resendFiles) {
/**
* Load conversation files for ALL agents, not just the initial agent.
* This enables handoff agents to access files that were uploaded earlier
* in the conversation. Without this, file_search and execute_code tools
* on handoff agents would fail to find previously attached files.
*/
if (conversationId != null && resendFiles) {
const fileIds = (await db.getConvoFiles(conversationId)) ?? [];
const toolResourceSet = new Set<EToolResources>();
for (const tool of agent.tools ?? []) {
@ -162,7 +168,7 @@ export async function initializeAgent(
if (requestFiles.length || toolFiles.length) {
currentFiles = (await db.updateFilesUsage(requestFiles.concat(toolFiles))) as IMongoFile[];
}
} else if (isInitialAgent && requestFiles.length) {
} else if (requestFiles.length) {
currentFiles = (await db.updateFilesUsage(requestFiles)) as IMongoFile[];
}