Agent Studio

Current Roadmap

MVP development progress and milestones

Current Roadmap

Track the development progress of Agent Studio MVP.

Phase 1: Foundation ✅

TaskStatus
Initialize repository with uv✅ Done
Configure pyproject.toml with dependencies✅ Done
Set up project structure✅ Done
Create Pydantic Settings✅ Done
Create exception hierarchy✅ Done
Create database engine & session✅ Done

Phase 2: Database Layer ✅

TaskStatus
Tenant model✅ Done
Agent model (with latest provider versions)✅ Done
Workflow model (DAG-based)✅ Done
Tool model (declarative actions)✅ Done
ProviderKey model (BYOK)✅ Done
APIKey model (scoped permissions)✅ Done
Call model (with transcripts & metrics)✅ Done
Base repository (generic CRUD)✅ Done
TenantScopedRepository✅ Done
Tenant repository✅ Done
Agent repository✅ Done
Workflow repository✅ Done
Alembic migrations setup✅ Done

Provider Versions (January 2026)

ProviderModelNotes
DeepgramNova-36.84% streaming WER
CartesiaSonic-340ms TTFA, emotion support
Gemini2.5 Flash/Pro, 3.0 Flash/ProLatest models
OpenAIGPT-4.1300k context, 55% SWE-bench
Silerov5Low-latency VAD

Phase 3: API Layer ✅

TaskStatus
FastAPI application factory✅ Done
JWT authentication✅ Done
API key authentication✅ Done
Permission/scope checking✅ Done
Health router✅ Done
Request ID middleware✅ Done
Logging middleware✅ Done
Dependency injection (dependencies.py)✅ Done
Auth router✅ Done
Agents router✅ Done
Workflows router✅ Done
Tools router✅ Done
Calls router✅ Done
Tool repository✅ Done
Call repository✅ Done
Rate limiting middlewarePending

Phase 4: Provider System ✅

TaskStatus
Provider base interfaces (Protocol)✅ Done
Provider registry✅ Done
Deepgram STT (Nova-3)✅ Done
Sarvam STT (Saarika v2)✅ Done
Cartesia TTS (Sonic-3)✅ Done
Sarvam TTS (Bulbul v2)✅ Done
Gemini LLM (2.5/3.0)✅ Done
OpenAI LLM (GPT-4.1)✅ Done
Silero VAD (v5)✅ Done

Phase 5: Core Domain ✅

TaskStatus
CallContext manager✅ Done
Domain models (AgentConfig, WorkflowConfig, etc.)✅ Done
Variable resolver (template engine)✅ Done
Tool executor✅ Done
Action handlers (all types)✅ Done
Workflow runner✅ Done
Handoff logic✅ Done

CallContext Features

  • Dot-notation path access (user.name, workflow.logged_meals)
  • Template resolution ({{user.name}})
  • Array append operations (logged_meals[+])
  • Per-agent state isolation

Tool Action Types

  • context.set, context.get, context.delete
  • webhook (with endpoint resolution)
  • handoff, respond
  • conditional (with nested actions)
  • flag.set, flag.clear
  • validate, transform, log

Phase 6: Worker Process ✅

TaskStatus
LiveKit worker entrypoint✅ Done
Session management (CallSession)✅ Done
Internal API client (WorkerAPIClient)✅ Done
Provider factory (BYOK support)✅ Done
Endpoint resolver✅ Done
Transcript collection✅ Done

Provider Key Inheritance

Priority: Agent Override → Tenant BYOK → Platform Default

Phase 7: Documentation ✅

TaskStatus
Fumadocs setup✅ Done
Architecture docs✅ Done
Data model docs✅ Done
ADRs (001-006)✅ Done
Getting started (installation)✅ Done
Concept docs (agents, workflows, tools, tenants)✅ Done
Roadmap updates✅ Done
API reference✅ Done
Deployment guides✅ Done

Phase 8: Testing & CI/CD ✅

TaskStatus
pytest configuration✅ Done
Test fixtures (conftest.py)✅ Done
Unit tests - Core (106 tests)✅ Done
Unit tests - Providers (19 tests)✅ Done
Unit tests - API (20 tests)✅ Done
Integration tests (calls)✅ Done
Testcontainers setup (PostgreSQL)✅ Done
Testing documentation✅ Done
GitHub Actions CI✅ Done
Docker build pipeline✅ Done

Test Summary

  • Unit Tests: 145 passing
  • Integration Tests: 13 passing (PostgreSQL via testcontainers)
  • Type Checking: 0 errors, 4 warnings (optional deps)

CI/CD Infrastructure

  • GitHub Actions: Lint, test, build, Docker image
  • Docker: Production multi-stage build, dev with hot reload
  • docker-compose: Full stack with PostgreSQL, Redis, LiveKit

Type Safety

The codebase passes strict type checking with basedpyright:

  • 0 errors across all modules
  • 4 warnings (all for optional/dynamic imports)
  • Proper Protocol-based interfaces
  • Full type annotations

Milestones

M1: Foundation Complete ✅

  • All Phase 1 & 2 tasks done
  • Database models and repositories working

M2: API & Providers Complete ✅

  • Phases 3 & 4 done
  • Full authentication system
  • All voice providers integrated

M3: Voice Pipeline Complete ✅

  • Phases 5 & 6 done
  • CallContext with template resolution
  • Declarative tool system
  • Multi-agent workflow orchestration
  • LiveKit worker integration

M4: Backend MVP Ready ✅

  • Documentation complete
  • Tests passing (unit + integration)
  • CI/CD pipeline
  • Alembic migrations setup

M5: Full MVP with Dashboard ✅

  • Dashboard UI complete
  • Workflow visual builder
  • Observability & analytics
  • End-to-end user flows

Phase 9: Dashboard (Next.js) 🔄

The dashboard is a Next.js application in apps/dashboard/ providing the UI for managing agents, workflows, and monitoring calls.

9.1 Project Setup ✅

TaskStatus
Initialize Next.js 16 with App Router✅ Done
Configure TypeScript 5.9✅ Done
Set up Tailwind CSS v4✅ Done
Add shadcn/ui components (23 components)✅ Done
Configure pnpm workspace✅ Done
Set up environment variables✅ Done

9.2 Authentication & Layout ✅

TaskStatus
Auth pages (login, register)✅ Done
JWT token management✅ Done
Protected route middleware✅ Done
Dashboard layout (sidebar, header)✅ Done
Auth context provider✅ Done
Dark mode support✅ Done

9.3 API Integration ✅

TaskStatus
TypeScript API types✅ Done
API client setup (fetch)✅ Done
React Query for data fetching✅ Done
Error handling & toast notifications✅ Done
Optimistic updatesPending

9.4 Core Features ✅

TaskStatus
Settings page✅ Done
Provider keys management (BYOK)✅ Done
API keys management page✅ Done
User/team managementPending

9.5 Agent Management ✅

TaskStatus
Agents list view✅ Done
Agent create/edit form✅ Done
Prompt editor with variables✅ Done
Provider configuration (STT/TTS/LLM)✅ Done
Tool assignment✅ Done
Agent testing/previewPending

9.6 Workflow Builder ✅

TaskStatus
Workflow list and forms✅ Done
Visual DAG editor (React Flow)✅ Done
Node palette (drag-drop agents)✅ Done
Connection management✅ Done
Skip condition editor✅ Done
Context passing configuration✅ Done
Workflow validation✅ Done
Workflow testing✅ Done

9.7 Tool Management ✅

TaskStatus
Tools list view✅ Done
Tool create/edit form✅ Done
Parameter schema editor✅ Done
Action chain builder✅ Done
Conditional action editor✅ Done
Webhook endpoint configuration✅ Done

9.8 Call Monitoring ✅

TaskStatus
Calls list view✅ Done
Call detail view✅ Done
Transcript viewer (search, filter, export)✅ Done
Agent flow visualization✅ Done
Analytics dashboard✅ Done
Call metrics & analytics✅ Done

9.9 Observability ✅

TaskStatus
Real-time call events (WebSocket)✅ Done
Error logs viewer✅ Done
Usage metrics dashboard✅ Done
Provider latency charts✅ Done

Dashboard Technology Stack

LayerTechnologyPurpose
FrameworkNext.js 16 (App Router)React framework
LanguageTypeScript 5.9Type safety
StylingTailwind CSS v4Utility-first CSS
Componentsshadcn/uiAccessible components
StateReact QueryServer state
FormsReact Hook Form + ZodForm validation
Workflow EditorReact FlowVisual DAG builder
ChartsRechartsMetrics visualization
IconsLucide ReactIcon library

Dashboard Pages Created

PageRouteStatus
Home/✅ Done
Login/login✅ Done
Register/register✅ Done
Dashboard/dashboard✅ Done
Agents/dashboard/agents✅ Done
Workflows/dashboard/workflows✅ Done
Tools/dashboard/tools✅ Done
Calls/dashboard/calls✅ Done
Analytics/dashboard/analytics✅ Done
Settings/dashboard/settings✅ Done
API Keys/dashboard/settings/api-keys✅ Done
Provider Keys/dashboard/settings/providers✅ Done
New Agent/dashboard/agents/new✅ Done
Edit Agent/dashboard/agents/[name]✅ Done
New Workflow/dashboard/workflows/new✅ Done
Edit Workflow/dashboard/workflows/[slug]✅ Done
New Tool/dashboard/tools/new✅ Done
Edit Tool/dashboard/tools/[name]✅ Done
Call Detail/dashboard/calls/[id]✅ Done
Workflow Editor/dashboard/workflows/[slug]/editor✅ Done
Workflow Test/dashboard/workflows/[slug]/test✅ Done
Webhook Endpoints/dashboard/settings/endpoints✅ Done

Phase 10: Production & Integration ✅

Production-ready features, backend integration, and operational improvements.

10.1 Webhook System ✅

TaskStatus
Webhook event types (call lifecycle)✅ Done
HMAC-SHA256 signature verification✅ Done
Exponential backoff retries (5 attempts)✅ Done
Webhook configuration per tenant✅ Done
Event filtering (subscribe to specific events)✅ Done
Webhook management API endpoints✅ Done
Secret rotation support✅ Done
Test webhook endpoint✅ Done
ADR-011 Webhook Delivery System✅ Done

Webhook Events

  • call.started - Call initiated
  • call.connected - Call connected to user
  • call.completed - Call ended normally
  • call.failed - Call failed
  • call.disconnected - Call dropped unexpectedly
  • call.timeout - Call timed out
  • call.agent.changed - Agent handoff occurred
  • call.transcript.updated - Transcript updated (optional)

10.2 Production Deployment ✅

TaskStatus
Production Docker Compose file✅ Done
Environment template (.env.production.template)✅ Done
Multi-replica API/Worker scaling✅ Done
Health check configuration✅ Done
ADR-009 Event-Driven Integration✅ Done
ADR-010 Production Deployment Strategy✅ Done

10.3 Backend Integration ✅

TaskStatus
Backend integration guide✅ Done
Kafka consumer example✅ Done
Python integration client✅ Done
API key scopes for integration✅ Done

10.4 Analytics & Observability Fixes ✅

TaskStatus
Fix orphaned call detection✅ Done
Active calls time-bounded query✅ Done
Cleanup orphaned calls endpoint✅ Done
Call status enum alignment (API/DB/Frontend)✅ Done
Call list null workflow handling✅ Done

10.5 Single Agent Testing ✅

TaskStatus
Agent test dispatch endpoint✅ Done
Ephemeral single-agent workflow✅ Done
Test mode metadata tracking✅ Done

Future Phases

Phase 11: Advanced Features (Planned)

TaskStatus
Rate limiting middlewarePending
Optimistic updates (dashboard)Pending
User/team managementPending
Agent testing/preview in dashboardPending
Multi-region deploymentPending
Kubernetes deployment guidePending
Per-agent worker scalingPending

Phase 12: Enterprise Features (Planned)

TaskStatus
SSO/SAML integrationPending
Audit loggingPending
Custom brandingPending
SLA monitoringPending
Advanced analyticsPending

Testing

Guide to testing Agent Studio - unit tests, integration tests, and test infrastructure

Architecture Decision Records

Key architectural decisions and their rationale

On this page

Current RoadmapPhase 1: Foundation ✅Phase 2: Database Layer ✅Provider Versions (January 2026)Phase 3: API Layer ✅Phase 4: Provider System ✅Phase 5: Core Domain ✅CallContext FeaturesTool Action TypesPhase 6: Worker Process ✅Provider Key InheritancePhase 7: Documentation ✅Phase 8: Testing & CI/CD ✅Test SummaryCI/CD InfrastructureType SafetyMilestonesM1: Foundation Complete ✅M2: API & Providers Complete ✅M3: Voice Pipeline Complete ✅M4: Backend MVP Ready ✅M5: Full MVP with Dashboard ✅Phase 9: Dashboard (Next.js) 🔄9.1 Project Setup ✅9.2 Authentication & Layout ✅9.3 API Integration ✅9.4 Core Features ✅9.5 Agent Management ✅9.6 Workflow Builder ✅9.7 Tool Management ✅9.8 Call Monitoring ✅9.9 Observability ✅Dashboard Technology StackDashboard Pages CreatedPhase 10: Production & Integration ✅10.1 Webhook System ✅Webhook Events10.2 Production Deployment ✅10.3 Backend Integration ✅10.4 Analytics & Observability Fixes ✅10.5 Single Agent Testing ✅Future PhasesPhase 11: Advanced Features (Planned)Phase 12: Enterprise Features (Planned)