EverMemoryArchive

Appearance

ema / ema / ContextManager

Class: ContextManager ​

Defined in: agent.ts:160

Manages conversation context and message history for the agent.

Constructors ​

Constructor ​

ts
new ContextManager(
   systemPrompt, 
   llmClient, 
   tools, 
   workspaceDir, 
   tokenLimit, 
   events): ContextManager;

Defined in: agent.ts:172

Parameters ​

systemPrompt ​

string

llmClient ​

LLMClient

tools ​

Tool[]

workspaceDir ​

string

tokenLimit ​

number = 80000

events ​

AgentEventsEmitter

Returns ​

ContextManager

Properties ​

apiTotalTokens ​

ts
apiTotalTokens: number;

Defined in: agent.ts:169

events ​

ts
events: AgentEventsEmitter;

Defined in: agent.ts:165

llmClient ​

ts
llmClient: LLMClient;

Defined in: agent.ts:161

messages ​

ts
messages: Message[];

Defined in: agent.ts:168

skipNextTokenCheck ​

ts
skipNextTokenCheck: boolean;

Defined in: agent.ts:170

systemPrompt ​

ts
systemPrompt: string;

Defined in: agent.ts:163

tokenLimit ​

ts
tokenLimit: number;

Defined in: agent.ts:164

toolDict ​

ts
toolDict: Map<string, Tool>;

Defined in: agent.ts:167

tools ​

ts
tools: Tool[];

Defined in: agent.ts:166

workspaceDir ​

ts
workspaceDir: string;

Defined in: agent.ts:162

Accessors ​

context ​

Get Signature ​

ts
get context(): Context;

Defined in: agent.ts:209

Get current conversation context (messages and tools).

Returns ​

Context

Methods ​

addAssistantMessage() ​

ts
addAssistantMessage(response): void;

Defined in: agent.ts:219

Add an assistant message to context.

Parameters ​

response ​

LLMResponse

Returns ​

void

addToolMessage() ​

ts
addToolMessage(
   result, 
   toolCallId, 
   name): void;

Defined in: agent.ts:229

Add a tool result message to context.

Parameters ​

result ​

ToolResult

toolCallId ​

string

name ​

string

Returns ​

void

addUserMessage() ​

ts
addUserMessage(content): void;

Defined in: agent.ts:214

Add a user message to context.

Parameters ​

content ​

string

Returns ​

void

createSummary() ​

ts
createSummary(messages, roundNum): Promise<string>;

Defined in: agent.ts:448

Create summary for one execution round.

Parameters ​

messages ​

Message[]

roundNum ​

number

Returns ​

Promise<string>

estimateTokens() ​

ts
estimateTokens(): number;

Defined in: agent.ts:247

Accurately calculate token count for message history using tiktoken.

Returns ​

number

estimateTokensFallback() ​

ts
estimateTokensFallback(): number;

Defined in: agent.ts:289

Fallback token estimation method (when tiktoken is unavailable).

Returns ​

number

getHistory() ​

ts
getHistory(): Message[];

Defined in: agent.ts:516

Get message history (shallow copy).

Returns ​

Message[]

summarizeMessages() ​

ts
summarizeMessages(): Promise<void>;

Defined in: agent.ts:328

Check and summarize message history if token limit exceeded.

Strategy (Agent mode):

  • Keep all user messages (these are user intents)
  • Summarize content between each user-user pair (agent execution process)
  • If last round is still executing (has agent/tool messages but no next user), also summarize
  • Structure: system -> user1 -> summary1 -> user2 -> summary2 -> user3 -> summary3 (if executing)

Summary is triggered when EITHER:

  • Local token estimation exceeds limit
  • API reported total_tokens exceeds limit

Returns ​

Promise<void>

updateApiTokens() ​

ts
updateApiTokens(response): void;

Defined in: agent.ts:240

Update API reported token count.

Parameters ​

response ​

LLMResponse

Returns ​

void