[Roadmap] Build QuickVoice into a comprehensive AI voice-agent platform #76

Vision

Build QuickVoice into a comprehensive, self-hostable, provider-neutral voice-agent platform that is approachable for SMBs and entrepreneurs while meeting the reliability, governance, extensibility, and data-control needs of enterprises and managed-service partners.

This roadmap is grounded in the current repository rather than treating the product as a blank slate. QuickVoice already has configurable agents, phone numbers, inbound and outbound calls, batch campaigns, call logs and recordings, live transcripts, knowledge ingestion, HTTP tools, an MCP marketplace and MCP server, secrets, API keys and OpenAPI docs, widgets, basic analytics, per-call data extraction/evaluation, audit records, organization roles, retention controls, and provider integrations. The work below extends and connects those capabilities.

Product principles

• Self-hostable and inspectable: every critical runtime and data path remains reviewable and operable by the customer.
• No-code approachable, API-first underneath: an SMB can launch from a template while an engineering team can automate the same resources.
• Provider-neutral: telephony, STT, LLM, TTS, storage, observability, and integration providers are replaceable policy choices.
• Safe by default: consent, least privilege, secret isolation, idempotency, auditability, and retention are platform primitives.
• Lifecycle complete: users can build, test, deploy, operate, analyze, and improve agents without leaving the product.
• Evidence over claims: compliance and reliability surfaces expose controls and evidence without claiming certifications the project has not earned.

Primary users

• Small businesses launching a receptionist, scheduler, support, lead-qualification, reminder, or follow-up agent.
• Entrepreneurs and agencies building reusable solutions for multiple clients.
• Operations teams running outbound campaigns and reviewing outcomes.
• Developers embedding voice or controlling agents through APIs, SDKs, webhooks, MCP, and infrastructure as code.
• Enterprise platform, security, compliance, and support teams governing production deployments.

Success measures

• Median time from organization creation to a successful preview and first real call.
• Template activation and deployment completion rates.
• Task success, containment, conversion, transfer, abandonment, and callback rates.
• End-to-end and component latency, interruption quality, silence, and tool success.
• Cost per call, minute, completed task, qualified lead, and conversion.
• Evaluation pass rate, flakiness, escaped regression rate, and human-review agreement.
• Campaign contact, answer, voicemail, suppression, retry, and conversion outcomes.
• Runtime availability, provider failover success, queue delay, and recovery time.
• Enterprise readiness: scoped access, approval coverage, audit completeness, retention enforcement, and exportability.

Delivery phases

P0 — Shared foundations

• Productionize the existing flow-builder work and define workflow/version/deployment contracts.
• Establish canonical events, metrics, costs, latency spans, outcomes, and customer/consent records.
• Stabilize public API conventions and event delivery.
• Define projects/environments, immutable versions, evaluation primitives, and safe promotion.

P1 — Complete the core product loop

• Visual workflow editor and durable runtime.
• Versioned templates and guided onboarding.
• Contacts, safe campaigns, advanced analytics, conversation intelligence, and automated evaluations.
• High-value integrations, transfers, callbacks, and operational diagnostics.

P2 — Scale, optimization, and ecosystem

• Experiments, canaries, advanced quality intelligence, enterprise identity and governance.
• BYOC/SIP and regional controls, warehouse exports, partner/reseller workflows, and multi-channel SDKs.

Epics

• [Epic] Visual orchestration, versions, and deployment #77
• [Epic] Templates, onboarding, and solution blueprints #78
• [Epic] Customer data and campaigns #79
• [Epic] Analytics, conversation intelligence, evaluations, and experiments #80
• [Epic] API, MCP, integrations, and data portability #81
• [Epic] Telephony, channels, and runtime reliability #82
• [Epic] Enterprise governance, security, privacy, and compliance #83
• [Epic] Platform UX, collaboration, commercial operations, and ecosystem #84

Cross-cutting acceptance criteria

• Every resource and query is organization scoped and authorized.
• Mutating public operations are idempotent where retries are plausible.
• Draft and production behavior are visibly separated; deployment is versioned and reversible.
• Zero-PII and retention modes are honored by logs, analytics, evaluations, exports, integrations, and observability.
• Provider-backed behavior exposes setup prerequisites, cost boundaries, degraded states, and actionable diagnostics.
• New UI is keyboard accessible, responsive, and does not rely on color alone.
• API and event contracts are documented and covered by contract tests.
• High-risk telephony, tool, migration, auth, billing, and privacy paths include failure and rollback tests.

Existing work to reconcile

• The unmerged agent-flow-builder-20260629 branch contains a substantial but stale workflow implementation and should be evaluated before new flow work begins.
• Several open Langfuse PRs overlap. Observability work must be consolidated into one reviewed implementation rather than duplicated.
• Existing basic campaigns, analytics, templates, MCP, API docs, widgets, secrets, and audit capabilities should be migrated or extended in place.

Out of scope for this roadmap issue

This issue coordinates product work; it does not itself authorize a broad implementation PR. Each linked child issue defines a reviewable outcome, boundaries, dependencies, tests, and rollout plan.