🤖 feat: Agent Handoffs (Routing) (#10176)

* feat: Add support for agent handoffs with edges in agent forms and schemas

chore: Mark `agent_ids` field as deprecated in favor of edges across various schemas and types

chore: Update dependencies for @langchain/core and @librechat/agents to latest versions

chore: Update peer dependency for @librechat/agents to version 3.0.0-rc2 in package.json

chore: Update @librechat/agents dependency to version 3.0.0-rc3 in package.json and package-lock.json

feat: first pass, multi-agent handoffs

fix: update output type to ToolMessage in memory handling functions

fix: improve type checking for graphConfig in createRun function

refactor: remove unused content filtering logic in AgentClient

chore: update @librechat/agents dependency to version 3.0.0-rc4 in package.json and package-lock.json

fix: update @langchain/core peer dependency version to ^0.3.72 in package.json and package-lock.json

fix: update @librechat/agents dependency to version 3.0.0-rc6 in package.json and package-lock.json; refactor stream rate handling in various endpoints

feat: Agent handoff UI

chore: update @librechat/agents dependency to version 3.0.0-rc8 in package.json and package-lock.json

fix: improve hasInfo condition and adjust UI element classes in AgentHandoff component

refactor: remove current fixed agent display from AgentHandoffs component due to redundancy

feat: enhance AgentHandoffs UI with localized beta label and improved layout

chore: update @librechat/agents dependency to version 3.0.0-rc10 in package.json and package-lock.json

feat: add `createSequentialChainEdges` function to add back agent chaining via multi-agents

feat: update `createSequentialChainEdges` call to only provide conversation context between agents

feat: deprecate Agent Chain functionality and update related methods for improved clarity

* chore: update @librechat/agents dependency to version 3.0.0-rc11 in package.json and package-lock.json

* refactor: remove unused addCacheControl function and related imports and import from @librechat/agents

* chore: remove unused i18n keys

* refactor: remove unused format export from index.ts

* chore: update @librechat/agents to v3.0.0-rc13

* chore: remove BEDROCK_LEGACY provider from Providers enum

* chore: update @librechat/agents to version 3.0.2 in package.json
This commit is contained in:
Danny Avila 2025-11-05 17:15:17 -05:00 committed by GitHub
parent 958a6c7872
commit 8a4a5a4790
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
41 changed files with 1108 additions and 3810 deletions

View file

@ -3,7 +3,6 @@ const { logger } = require('@librechat/data-schemas');
const { DynamicStructuredTool } = require('@langchain/core/tools');
const { getBufferString, HumanMessage } = require('@langchain/core/messages');
const {
sendEvent,
createRun,
Tokenizer,
checkAccess,
@ -12,14 +11,12 @@ const {
resolveHeaders,
getBalanceConfig,
memoryInstructions,
formatContentStrings,
getTransactionsConfig,
createMemoryProcessor,
} = require('@librechat/api');
const {
Callback,
Providers,
GraphEvents,
TitleMethod,
formatMessage,
formatAgentMessages,
@ -38,12 +35,12 @@ const {
bedrockInputSchema,
removeNullishValues,
} = require('librechat-data-provider');
const { addCacheControl, createContextHandlers } = require('~/app/clients/prompts');
const { initializeAgent } = require('~/server/services/Endpoints/agents/agent');
const { spendTokens, spendStructuredTokens } = require('~/models/spendTokens');
const { getFormattedMemories, deleteMemory, setMemory } = require('~/models');
const { encodeAndFormat } = require('~/server/services/Files/images/encode');
const { getProviderConfig } = require('~/server/services/Endpoints');
const { createContextHandlers } = require('~/app/clients/prompts');
const { checkCapability } = require('~/server/services/Config');
const BaseClient = require('~/app/clients/BaseClient');
const { getRoleByName } = require('~/models/Role');
@ -80,8 +77,6 @@ const payloadParser = ({ req, agent, endpoint }) => {
return req.body.endpointOption.model_parameters;
};
const noSystemModelRegex = [/\b(o1-preview|o1-mini|amazon\.titan-text)\b/gi];
function createTokenCounter(encoding) {
return function (message) {
const countTokens = (text) => Tokenizer.getTokenCount(text, encoding);
@ -803,137 +798,81 @@ class AgentClient extends BaseClient {
);
/**
*
* @param {Agent} agent
* @param {BaseMessage[]} messages
* @param {number} [i]
* @param {TMessageContentParts[]} [contentData]
* @param {Record<string, number>} [currentIndexCountMap]
*/
const runAgent = async (agent, _messages, i = 0, contentData = [], _currentIndexCountMap) => {
config.configurable.model = agent.model_parameters.model;
const currentIndexCountMap = _currentIndexCountMap ?? indexTokenCountMap;
if (i > 0) {
this.model = agent.model_parameters.model;
const runAgents = async (messages) => {
const agents = [this.options.agent];
if (
this.agentConfigs &&
this.agentConfigs.size > 0 &&
((this.options.agent.edges?.length ?? 0) > 0 ||
(await checkCapability(this.options.req, AgentCapabilities.chain)))
) {
agents.push(...this.agentConfigs.values());
}
if (i > 0 && config.signal == null) {
config.signal = abortController.signal;
}
if (agent.recursion_limit && typeof agent.recursion_limit === 'number') {
config.recursionLimit = agent.recursion_limit;
if (agents[0].recursion_limit && typeof agents[0].recursion_limit === 'number') {
config.recursionLimit = agents[0].recursion_limit;
}
if (
agentsEConfig?.maxRecursionLimit &&
config.recursionLimit > agentsEConfig?.maxRecursionLimit
) {
config.recursionLimit = agentsEConfig?.maxRecursionLimit;
}
config.configurable.agent_id = agent.id;
config.configurable.name = agent.name;
config.configurable.agent_index = i;
const noSystemMessages = noSystemModelRegex.some((regex) =>
agent.model_parameters.model.match(regex),
);
const systemMessage = Object.values(agent.toolContextMap ?? {})
.join('\n')
.trim();
// TODO: needs to be added as part of AgentContext initialization
// const noSystemModelRegex = [/\b(o1-preview|o1-mini|amazon\.titan-text)\b/gi];
// const noSystemMessages = noSystemModelRegex.some((regex) =>
// agent.model_parameters.model.match(regex),
// );
// if (noSystemMessages === true && systemContent?.length) {
// const latestMessageContent = _messages.pop().content;
// if (typeof latestMessageContent !== 'string') {
// latestMessageContent[0].text = [systemContent, latestMessageContent[0].text].join('\n');
// _messages.push(new HumanMessage({ content: latestMessageContent }));
// } else {
// const text = [systemContent, latestMessageContent].join('\n');
// _messages.push(new HumanMessage(text));
// }
// }
// let messages = _messages;
// if (agent.useLegacyContent === true) {
// messages = formatContentStrings(messages);
// }
// if (
// agent.model_parameters?.clientOptions?.defaultHeaders?.['anthropic-beta']?.includes(
// 'prompt-caching',
// )
// ) {
// messages = addCacheControl(messages);
// }
let systemContent = [
systemMessage,
agent.instructions ?? '',
i !== 0 ? (agent.additional_instructions ?? '') : '',
]
.join('\n')
.trim();
if (noSystemMessages === true) {
agent.instructions = undefined;
agent.additional_instructions = undefined;
} else {
agent.instructions = systemContent;
agent.additional_instructions = undefined;
}
if (noSystemMessages === true && systemContent?.length) {
const latestMessageContent = _messages.pop().content;
if (typeof latestMessageContent !== 'string') {
latestMessageContent[0].text = [systemContent, latestMessageContent[0].text].join('\n');
_messages.push(new HumanMessage({ content: latestMessageContent }));
} else {
const text = [systemContent, latestMessageContent].join('\n');
_messages.push(new HumanMessage(text));
}
}
let messages = _messages;
if (agent.useLegacyContent === true) {
messages = formatContentStrings(messages);
}
const defaultHeaders =
agent.model_parameters?.clientOptions?.defaultHeaders ??
agent.model_parameters?.configuration?.defaultHeaders;
if (defaultHeaders?.['anthropic-beta']?.includes('prompt-caching')) {
messages = addCacheControl(messages);
}
if (i === 0) {
memoryPromise = this.runMemory(messages);
}
/** Resolve request-based headers for Custom Endpoints. Note: if this is added to
* non-custom endpoints, needs consideration of varying provider header configs.
*/
if (agent.model_parameters?.configuration?.defaultHeaders != null) {
agent.model_parameters.configuration.defaultHeaders = resolveHeaders({
headers: agent.model_parameters.configuration.defaultHeaders,
body: config.configurable.requestBody,
});
}
memoryPromise = this.runMemory(messages);
run = await createRun({
agent,
req: this.options.req,
agents,
indexTokenCountMap,
runId: this.responseMessageId,
signal: abortController.signal,
customHandlers: this.options.eventHandlers,
requestBody: config.configurable.requestBody,
tokenCounter: createTokenCounter(this.getEncoding()),
});
if (!run) {
throw new Error('Failed to create run');
}
if (i === 0) {
this.run = run;
}
if (contentData.length) {
const agentUpdate = {
type: ContentTypes.AGENT_UPDATE,
[ContentTypes.AGENT_UPDATE]: {
index: contentData.length,
runId: this.responseMessageId,
agentId: agent.id,
},
};
const streamData = {
event: GraphEvents.ON_AGENT_UPDATE,
data: agentUpdate,
};
this.options.aggregateContent(streamData);
sendEvent(this.options.res, streamData);
contentData.push(agentUpdate);
run.Graph.contentData = contentData;
}
this.run = run;
if (userMCPAuthMap != null) {
config.configurable.userMCPAuthMap = userMCPAuthMap;
}
/** @deprecated Agent Chain */
config.configurable.last_agent_id = agents[agents.length - 1].id;
await run.processStream({ messages }, config, {
keepContent: i !== 0,
tokenCounter: createTokenCounter(this.getEncoding()),
indexTokenCountMap: currentIndexCountMap,
maxContextTokens: agent.maxContextTokens,
callbacks: {
[Callback.TOOL_ERROR]: logToolError,
},
@ -942,109 +881,22 @@ class AgentClient extends BaseClient {
config.signal = null;
};
await runAgent(this.options.agent, initialMessages);
let finalContentStart = 0;
if (
this.agentConfigs &&
this.agentConfigs.size > 0 &&
(await checkCapability(this.options.req, AgentCapabilities.chain))
) {
const windowSize = 5;
let latestMessage = initialMessages.pop().content;
if (typeof latestMessage !== 'string') {
latestMessage = latestMessage[0].text;
}
let i = 1;
let runMessages = [];
const windowIndexCountMap = {};
const windowMessages = initialMessages.slice(-windowSize);
let currentIndex = 4;
for (let i = initialMessages.length - 1; i >= 0; i--) {
windowIndexCountMap[currentIndex] = indexTokenCountMap[i];
currentIndex--;
if (currentIndex < 0) {
break;
}
}
const encoding = this.getEncoding();
const tokenCounter = createTokenCounter(encoding);
for (const [agentId, agent] of this.agentConfigs) {
if (abortController.signal.aborted === true) {
break;
}
const currentRun = await run;
if (
i === this.agentConfigs.size &&
config.configurable.hide_sequential_outputs === true
) {
const content = this.contentParts.filter(
(part) => part.type === ContentTypes.TOOL_CALL,
);
this.options.res.write(
`event: message\ndata: ${JSON.stringify({
event: 'on_content_update',
data: {
runId: this.responseMessageId,
content,
},
})}\n\n`,
);
}
const _runMessages = currentRun.Graph.getRunMessages();
finalContentStart = this.contentParts.length;
runMessages = runMessages.concat(_runMessages);
const contentData = currentRun.Graph.contentData.slice();
const bufferString = getBufferString([new HumanMessage(latestMessage), ...runMessages]);
if (i === this.agentConfigs.size) {
logger.debug(`SEQUENTIAL AGENTS: Last buffer string:\n${bufferString}`);
}
try {
const contextMessages = [];
const runIndexCountMap = {};
for (let i = 0; i < windowMessages.length; i++) {
const message = windowMessages[i];
const messageType = message._getType();
if (
(!agent.tools || agent.tools.length === 0) &&
(messageType === 'tool' || (message.tool_calls?.length ?? 0) > 0)
) {
continue;
}
runIndexCountMap[contextMessages.length] = windowIndexCountMap[i];
contextMessages.push(message);
}
const bufferMessage = new HumanMessage(bufferString);
runIndexCountMap[contextMessages.length] = tokenCounter(bufferMessage);
const currentMessages = [...contextMessages, bufferMessage];
await runAgent(agent, currentMessages, i, contentData, runIndexCountMap);
} catch (err) {
logger.error(
`[api/server/controllers/agents/client.js #chatCompletion] Error running agent ${agentId} (${i})`,
err,
);
}
i++;
}
await runAgents(initialMessages);
/** @deprecated Agent Chain */
if (config.configurable.hide_sequential_outputs) {
this.contentParts = this.contentParts.filter((part, index) => {
// Include parts that are either:
// 1. At or after the finalContentStart index
// 2. Of type tool_call
// 3. Have tool_call_ids property
return (
index >= this.contentParts.length - 1 ||
part.type === ContentTypes.TOOL_CALL ||
part.tool_call_ids
);
});
}
/** Note: not implemented */
if (config.configurable.hide_sequential_outputs !== true) {
finalContentStart = 0;
}
this.contentParts = this.contentParts.filter((part, index) => {
// Include parts that are either:
// 1. At or after the finalContentStart index
// 2. Of type tool_call
// 3. Have tool_call_ids property
return (
index >= finalContentStart || part.type === ContentTypes.TOOL_CALL || part.tool_call_ids
);
});
try {
const attachments = await this.awaitMemoryWithTimeout(memoryPromise);
if (attachments && attachments.length > 0) {