Skip to content

Where we are today & roadmap

Frontier is an AI sales-call coaching platform designed to give real-time guidance to sales representatives during live calls. This page outlines our current development focus, near-term product priorities, and strategic investment areas, providing an honest view of our progress and future direction, sourced directly from our project tracking systems.

Our engineering efforts are actively engaged in several key areas to refine the Frontier platform and enhance user experience. These projects represent the core of our ongoing development:

ProjectThemeNotes
Remove blockers for new usersActivationAddressing anything that currently blocks new users from adopting Frontier.
ActivationGrowthEnhancing the onboarding and initial experience for new users.
Desktop app + HUDProductDeveloping the core Electron-based desktop application, including the heads-up display (HUD) and main call window.
Code qualityPlatformMaintaining high standards for code quality, targeting A-grade scores for tools like Sonar and GitHub Security.
Test CoveragePlatformIncreasing unit and integration test coverage using vitest.
InfraPlatformImproving our deployment processes, logging, and overall observability infrastructure.
StabilityPlatformFocusing on bug fixes, infrastructure improvements, and enhanced validation to ensure a robust platform.

We are actively planning and executing on features that will significantly expand Frontier’s reach and capabilities for sales teams.

Frontier is currently exclusively available as a desktop application for Mac. To address the significant portion of sales teams that use Windows, developing a Windows version is a high commercial priority.

We are in the process of migrating how representative speech is transcribed. Currently, transcription often relies on third-party meeting capture services like Recall.ai. Our goal is to move to a system where representative speech is transcribed directly from the Frontier HUD window using Deepgram’s streaming model. This change is designed to improve the accuracy of speaker attribution and reduce transcription lag.

We are working to replace our current billing stubs with a robust payment system using Stripe. This will enable seat-based invoicing, a customer portal, and feature gating based on subscription plans. This integration is not yet live.

To provide sales representatives with critical context and streamline their workflows, we are building integrations with customer relationship management (CRM) systems. Composio is already integrated within our call server for CRM tool actions. Our initial focus is on Attio, with ongoing evaluation (spikes) between Composio and Merge.dev for the optimal integration layer.

We are developing an internal admin area to give organizations better control over their Frontier usage. This dashboard will initially focus on managing organizations, subscriptions, and platform health, starting with billing administration.

Beyond real-time coaching, we are enhancing the post-call experience. This includes generating better call summaries, facilitating more effective call reviews, and enabling seamless synchronization of call data back into CRM systems.

We are developing capabilities for sales playbooks that adapt in real-time. This includes personalizing scripts with variables auto-filled from calendar events or CRM data, and enabling conditional branching where the HUD shifts to “listening-for” a specific entity or phrase.

To provide sales leaders with actionable insights, we are building an administrative view of representative performance. This will include metrics such as script adherence and objection handling, leveraging the underlying knowledge graph.

To maintain our competitive edge and deliver increasing value, we are making strategic investments in core platform capabilities.

Our commitment to improving the quality and speed of AI coaching is ongoing. This includes developing robust evaluation pipelines, using tools like Promptfoo for model-speed evaluations, and optimizing the response time for signal detection and answer generation. We are also focused on improving FAQ extraction and post-call summary prompt quality. This work makes the computationally expensive aspects of AI-driven coaching more efficient.

The Knowledge Base is central to providing accurate and relevant real-time coaching. This is an area of active evaluation, and it is intentionally designed to be multi-backend to allow for continuous improvement and experimentation with different retrieval approaches.

Today, we utilize Cloudflare AI Search and Supermemory (which serves as a deliberate stop-gap solution). We are actively exploring and planning to experiment with Graph RAG (Retrieval Augmented Generation) services for the Knowledge Base layer, including options like Amazon Bedrock Knowledge Bases and Anthropic. Our sequencing prioritizes user experience first, then speed, with accuracy and completeness, particularly via Graph RAG and advanced knowledge-base modeling, as the next major investment area.

As we scale, ensuring robust compliance is paramount. Our roadmap includes achieving industry-standard certifications such as SOC-2 and adhering to data protection regulations like GDPR. This also covers user access and deletion (DSAR) requests, data sovereignty, and multi-region deployment capabilities.

We actively manage potential risks to ensure the Frontier platform remains reliable and secure.

We are focused on mitigating the risks associated with third-party service dependencies that are critical during live calls. This includes implementing an incident-communication banner, a provider health dashboard with an Inngest poller and Slack alerts, and building in regional routing and failover mechanisms for services like Recall.ai between US and EU regions. This directly addresses potential single points of failure from external providers during live calls.

Our agent-assisted development pipeline is designed for high throughput. We are actively hardening our agent workflow to guard against potential issues such as secret leakage, accidental production infrastructure destruction, and the execution of dangerous commands. This pairs with our automated review pipeline.

While our project roadmap is clear, certain quantitative metrics are not yet centrally tracked or verified in our codebase and require founder input.