Autonomous AI agents for ticket management, task execution, code generation, and strategic planning.

Agent SDK

The Agent SDK provides autonomous decision-making capabilities for AI agents. It ships four specialized agents, an intelligence layer for context analysis, and integrations for chat and CLI workflows.

npm install @codmir/sdk

import { createAgentSDK } from '@codmir/sdk/agent';

Quick Start

import { createAgentSDK } from '@codmir/sdk/agent';

const sdk = await createAgentSDK({
  autonomyMode: 'balanced',
  confidenceThreshold: 0.7,
  learningEnabled: true,
  contextDepth: 'deep',
  planningHorizon: 'tactical',
  apiEndpoint: process.env.CODMIR_API_URL,
  apiKey: process.env.CODMIR_API_KEY,
  projectId: 'project-123',
  features: {
    conversationIntelligence: true,
    codebaseAnalysis: true,
    automaticTaskCreation: true,
    proactiveTicketGeneration: true,
    strategicPlanning: true,
    learningFromFailures: true,
  },
  maxActionsPerMinute: 10,
  maxConcurrentTasks: 5,
  safetyChecksEnabled: true,
  humanApprovalRequired: ['deploy', 'delete', 'billing'],
  learning: {
    memoryRetentionDays: 30,
    patternRecognitionEnabled: true,
    successMetricsTracking: true,
    failureAnalysisEnabled: true,
  },
});

Autonomous Agents

The SDK includes four specialized agents, each designed for a specific domain.

TicketAgent

Creates, triages, and manages tickets based on conversation context and codebase analysis.

const ticket = await sdk.agents.ticketAgent.createTicket({
  context: conversationContext,
  triggerType: 'automatic',
});

TaskAgent

Breaks down objectives into subtasks, estimates effort, and tracks completion.

const task = await sdk.agents.taskAgent.createTask({
  objective: 'Implement OAuth2 flow',
  context: projectContext,
});

CodeAgent

Generates, reviews, and refactors code autonomously based on requirements.

const code = await sdk.generateCode(
  { description: 'Add rate limiting middleware', language: 'typescript' },
  { project: projectContext }
);

PlannerAgent

Creates strategic plans with milestones, dependencies, and risk assessments.

const plan = await sdk.createStrategicPlan(
  { goal: 'Migrate to microservices', timeline: '6 months' },
  { project: projectContext }
);

Intelligence Modules

Four intelligence modules analyze different aspects of your project to provide context for agent decisions.

Module	Purpose
ConversationIntelligence	Analyzes chat messages for intent, sentiment, topics, and entities
CodebaseIntelligence	Understands code structure, dependencies, and recent changes
TaskIntelligence	Tracks task patterns, estimates, and completion rates
ProjectIntelligence	Monitors project health, velocity, and strategic alignment

// Analyze a conversation for actionable insights
const analysis = await sdk.analyzeConversation(messages, context);

// Get cross-module insights
const insights = await sdk.getSDKInsights();
console.log(insights.intelligence.codebase);
console.log(insights.agents.ticket);

Integrations

Chat Integration

Connect the agent to your chat interface. Automatically processes messages, creates tickets, and generates tasks from conversations.

import { createChatAgent } from '@codmir/sdk/agent';

const chatAgent = await createChatAgent(config);

// Process a message through the agent
const response = await chatAgent.processMessage(
  'conv-123',
  { role: 'user', content: 'The login page is broken on mobile' },
  { project: projectContext }
);

// Create a ticket from the conversation
const ticket = await chatAgent.createTicket('conv-123', context);

// Get conversation insights
const insights = await chatAgent.getInsights('conv-123');

CLI Integration

Embed the agent in terminal workflows for automated task execution.

import { CLIIntegration, createCLIIntegration } from '@codmir/sdk/agent';

const cli = await createCLIIntegration(config);

Configuration

Autonomy Modes

Mode	Description
`conservative`	Always asks for approval before acting
`balanced`	Acts autonomously on low-risk decisions, asks for high-risk ones
`aggressive`	Acts autonomously on most decisions, only asks for critical ones
`autonomus`	Fully autonomous — no approval gates

Safety Controls

const config = {
  safetyChecksEnabled: true,
  humanApprovalRequired: ['deploy', 'delete', 'billing', 'merge'],
  maxActionsPerMinute: 10,
  maxConcurrentTasks: 5,
};

Actions listed in humanApprovalRequired always prompt for user confirmation, regardless of autonomy mode.

Core Engine

For advanced use cases, you can work directly with the AgentEngine and its modules.

import { AgentEngine, ContextManager, PlanningEngine, ActionExecutor } from '@codmir/sdk/agent';

const engine = new AgentEngine(config);
await engine.initialize();

const result = await engine.processInput({
  type: 'request',
  content: { objective: 'Fix failing CI pipeline' },
  context: projectContext,
});

console.log(result.decisions);
console.log(result.plans);

TypeScript Support

import type {
  AgentConfig,
  Context,
  ConversationIntent,
  CodebaseSnapshot,
  TicketContext,
  TaskContext,
} from '@codmir/sdk/agent';

Events

The agent system emits events you can subscribe to for monitoring and logging.

import { AgentEvents, IntegrationEvents } from '@codmir/sdk/agent';

AI Monitoring

Track LLM costs and performance for agent calls

Chat Client

Basic chat integration without agent autonomy

Serverless

Run agent workloads in the cloud