Roadmap
What we have built, and what comes next
Dialogist runs on its own stack: recognition, voice, knowledge, testing and telephony. Here is what has shipped, what the team is working on now and what is planned.
Shipped
Newest first.
Shipped: Oasis Speech 2
Recognition rebuilt for phone calls: an architectural change to the model, made for accuracy on short replies, narrowband audio and noisy lines. The agent now hears 48 languages, and every one of them improved.
Shipped: Audio pipeline optimization for low latency
From the caller’s last word to the agent’s first sound, the agent itself needs about a quarter of a second. Early start, dead air trimmed from synthesized speech and one media clock for all calls. The same work doubled the number of calls a Server Unit carries at Standard grade.
Shipped: Full Duplex Mode
The caller’s channel stays open. The agent hears from intonation when a turn is over, starts its reply before the pause ends, tells an interruption from a cough, lowers its voice before it stops and says “uh-huh” in the right place.
Shipped: Integration API
The agent works inside your systems: HTTP tools it calls during the conversation, signed webhooks with the result of every call, and connectors for Salesforce, Creatio and Zendesk.
Shipped: Oasis Speech 1
Our own speech recognition service, running on our servers and on yours. Recognition streams while the caller speaks, so only the last few tens of milliseconds of work remain when they stop.
Shipped: Polygon
The conversation test bench, in the Pro plan as Dialog simulation.
In progress
What the team is working on at the moment.
In progress: Language training
Continuous training of Oasis Speech across all 48 languages.
In progress: Turn-taking tuning
End-of-turn settings per language and per kind of agent.
In progress: Live compliance monitoring
Every live call is checked against your rules: required wording, forbidden phrases, topics for a person. A call that breaks a rule is flagged for review.
In progress: Agentic modes
Agents that plan and act over several steps within one call: look something up, decide, act in your system and confirm, within the same latency budget as a simple reply.
In progress: New Agentic Builder
A new way to build an agent inside Dialogist. You describe the job; the builder drafts the prompt, the tools, the data to extract and the tests, then rehearses the agent in Polygon before it goes live.
In progress: Statistics and analysis platform
Call analytics built for volume: outcomes, extracted fields, reasons for transfer and drop-off, conversation quality scores, and trends across agents, campaigns and languages.
In progress: Voicecore media core
The audio path of the voice engine, rewritten in Rust. Transport and signal processing are done; end-of-turn decisions in the core are planned.
Planned
Telephony
- Numbers and SIP trunks in the EUPlanned
- Skills-based routing and overflow chainsPlanned
Building agents
- A setup wizard inside DialogistPlanned
- A visual designer for custom toolsPlanned
Models
- Speech recognition fine-tuned on your dataPlanned
Security
- Enterprise SSO with SAMLPlanned
- Google and Apple sign-inPlanned
- More roles and custom rolesPlanned
- An automatic dedicated environment per organizationPlanned
Built in
- A minimal CRMPlanned
- A booking calendarPlanned
Connectors
- CRMPlanned
- Help deskPlanned
- Booking and calendarsPlanned
- Knowledge sourcesPlanned
Developers
- DocumentationPlanned
- MCP serverPlanned
Something missing?
Requests decide the order. Tell us what your team needs, and we will write when it ships.