Roadmap

What we've built and where we're going across the Omra platform — AI gateway, CRM, infrastructure, mail, and single sign-on.

52% 26 of 50 shipped
LLM · 10/23 CRM · 0/6 TOOLS · 5/7 MAIL · 7/9 SSO · 4/5

Now in progress

  1. CRM Phase 17 active right now

    Master Schedule

    The operator sets the master's work schedule — a weekly template and overrides for specific dates.

    • 17-01-PLAN.md — schema (break columns + `BookingScheduleOverride`) + migration + slot-calc priority chain (override > weekly, break carve) + unit tests (crit-4)
    • 17-02-PLAN.md — backend: weekly-preset bulk-upsert + override CRUD (tenant/IDOR guard, audit, conflict_count D-09) + integration tests
    • 17-03-PLAN.md — screen `/crm/booking/{page_id}/schedule` (presets + monthly grid + drill-in + conflict warning) + blocking human-verify for public slot-reflection
  2. LLM Phase 11 active right now

    New cabinet on a modern stack

    The current admin panel has been evolving for a year and has grown quite large — we're rebuilding it on a fresh, more performant stack. The goal is to significantly speed up new features and make maintenance easier. The old dashboard runs in parallel, pages are switched one by one, with no service downtime for clients.

    • We spun up a separate staging instance for safe experiments.
    • Laid the foundation: single sign-on, theming, shared component language
    • Migrated most pages to the new stack: pricing, release notes, email templates, AI models, services, integrations, OAuth apps, agents, SDK keys.
    • Knowledge base with Finder-style folder navigation — create, rename, cascade delete
    • New section — competitor price monitoring: the client adds their own site and competitor sites, the system automatically finds products and shows price discrepancies.
    • The core of the chat dashboard is already on the new stack: threads, message feed, background tasks, streaming responses.
    • Currently in progress: a unified brand style for the entire dashboard (light and dark theme).
    • Next up — the final migration of the chat cabinet (inspector, composer, files) and switching all clients over.

May 2026

  1. SSO Phase 1 19 May 2026

    Single sign-on for all Omra products

    We built our own authorization service: one account unlocks access to all platform products. Supports an international standard — you can connect any third-party app that works with open protocols.

    • Registration and login via email and password
    • Support for the OIDC standard — connecting third-party apps
    • Secure sessions with automatic expiration
    • Full audit of all inputs and outputs for security
    • Signed tokens with verifiable signature
  2. SSO Phase 2 19 May 2026

    Self-service account management

    Email confirmation, password reset, passwordless login via magic link. Sessions can be revoked at any time — if your phone gets stolen, access is terminated remotely.

    • Email confirmation via one-time link
    • Password reset via email without contacting support
    • Passwordless login via magic link
    • List of active sessions with the ability to revoke each one
  3. SSO Phase 3 19 May 2026

    Teams and roles within the organization

    Multiple users in one company, inviting new team members, flexible role system. Log out of all sessions with one button — for security during terminations.

    • Inviting colleagues via email
    • Roles: company owner, administrator, regular user
    • Switching between companies for agencies
    • Force logout from all company sessions
  4. SSO Phase 4 19 May 2026

    Login via Google, GitHub, and two-factor authentication

    You can log in through familiar services — Google, GitHub. Protection against password brute-forcing. An additional verification factor via an authenticator app.

    • Login via Google and GitHub without registration
    • Two-factor authentication via app
    • Backup codes in case you lose your phone
    • Protection against password brute-forcing by IP address
    • Connecting external applications via standard protocol
  5. MAIL Phase 7 18 May 2026

    Unified API for all Omra services

    All Omra products connect to email the same way — unified authorization, the same interface for creating mailboxes, aliases, and forwards.

    • Unified authorization protocol between services
    • Creating mailboxes and aliases from any Omra service
    • Managing redirects and auto-replies
    • Full action logging for audit
  6. TOOLS Phase 5 18 May 2026

    Orchestrator and connection to email infrastructure

    A management service that pings DNS, edge nodes, and email through a single API. Other Omra products are connected — they now talk to the infrastructure through one point.

    • Unified API for all infrastructure operations
    • Connecting to the mail infrastructure via service endpoints
    • Real-time availability checks for all nodes
    • Readiness to accept requests from Omra single sign-on
  7. TOOLS Phase 4 12 May 2026

    Automatic certificates and zone synchronization

    Certificates for client domains are issued and updated automatically on all edge points. Zones on backup DNS servers sync instantly after changes on the primary one.

    • Automatic Let's Encrypt issuance for any connected domain
    • Auto-renewal 30 days before expiration
    • Instant sync of zone changes to all slave servers
    • The readiness of the zone on the new slave server is checked automatically
  8. MAIL Phase 6 10 May 2026

    Trust from email providers

    Warming up IP addresses, handling bounces from Mail.ru and Yandex, and responding to user complaints. Deliverability is consistently above 95%.

    • Gradual warm-up of new IP addresses to full load
    • Connecting feedback loops with major providers
    • Auto-exclude complaining subscribers from mailings
    • Failure handling: temporary errors are retried, permanent ones are excluded
  9. LLM Phase 10 8 May 2026

    Voice POC: phone AI agent

    Breakthrough in voice channel: AI bot actually answers phone calls, understands live speech, and can hang up. Stereo recording for further analysis. Cheaper than any human operator.

    • Call to a test Argentine number → AI responds with the voice "Coral"
    • Recognizes Russian, Spanish, and English without switching.
    • Echo suppression on speakerphone, protection against false triggers from noise.
    • Stereo recording of each conversation (client on the left, bot on the right)
    • Cost ~$0.10 per minute of actual conversation

April 2026

  1. LLM Phase 9 27 Apr 2026

    Images, versions, Avito, new main website

    Several major parallel deliveries in a couple of days: image generation, updated main site, Avito integration, release history in the dashboard.

    • Image generation: Recraft V4, Ideogram V3, FLUX 2, GPT-Image 2 — each client with their own key
    • Integration with Avito: listings, bids, stats, chats — all available to the AI agent.
    • Main site redesigned: regional sections (Russia / Latin America / Global)
    • Release history: each updated component is visible in the dashboard with a description.
    • Build version and commit hash in the header — always clear what's in production
  2. TOOLS Phase 3 27 Apr 2026

    Internal network: everything resolves correctly

    Previously, services within the platform couldn't access their own public domains — traffic would go out and come back in. We set up internal resolution — now everything routes directly within the network.

    • Internal DNS server with automatic synchronization
    • 43 platform domains resolve instantly within the network
    • Container support: they also see internal addresses
    • External traffic goes through the external resolver by default.
  3. LLM Phase 5 21 Apr 2026

    Integrations for different markets

    We opened the platform for different geographies and industries. Argentine tax authority, Russian marketplaces, landing pages — each market requires its own integrations. Now the AI agent in the chat can "pay a tax" or "show orders from CDEK."

    • Argentina's tax authority (AFIP/ARCA): generating payment QR codes and statements
    • Class365 as an ERP engine for Russian sellers (orders, products, counterparties)
    • PromoVeer.ru — landing page with a request form, requests flow into the funnel automatically
    • Storage and delivery of payment receipts to customers
  4. LLM Phase 8 21 Apr 2026

    Email inside the platform

    Turned the platform into a full-fledged email client — clients no longer switch between our dashboard and Gmail. Incoming emails land in the chat, AI helps with drafts, templates get reused.

    • Incoming emails as threads right in the dashboard
    • "AI Draft" button — instant response sketch in the client's style
    • Email templates with preview and variables
    • Auto-forms: AI-generated landing pages get working forms without developer involvement
    • Static site deployment tools via our infrastructure (DNS + storage + acceleration)
  5. MAIL Phase 5 20 Apr 2026

    Email templates for AI agents

    An AI agent in the chat can ask to "send a client an email using the 'Greeting' template with these names and dates" — and the email goes out. Daily email limits per client.

    • Template storage with variables inside
    • The AI agent automatically fills data into the template.
    • Daily and per-minute email limits for each client
    • Logging every send to track delivery.
  6. LLM Phase 7 16 Apr 2026

    Platform health monitoring

    We consolidated all platform health metrics into a single screen and into Zabbix. If anything degrades — we get an alert before the client complains. The client's financial health is also visible: how much is left, how long it will last.

    • Summary screen: models, payment providers, task queues, balance
    • Zabbix integration: alerts for any component failure or slowdown
    • The "when will the money run out" forecast is calculated based on the average weekly spend.
    • Health-check endpoints for external monitors
  7. MAIL Phase 4 15 Apr 2026

    Reliable signatures for any domains

    Each client domain gets its own digital signature. Emails look legitimate to Gmail, Mail.ru, and Yandex — they don't end up in spam.

    • Unique signature for each connected domain
    • Automatic signing key rotation
    • Ability to sign emails from subdomains of the main domain
  8. LLM Phase 6 15 Apr 2026

    Billing and subscriptions in 3 currencies

    We built a full-fledged subscription with auto-renewal: the customer pays and gets access; if they don’t pay, they’re gently blocked until payment. Three payment providers, three currencies, and pricing plans.

    • Payments via YooKassa, Tinkoff, MercadoPago — Russian, Russian premium, and Latin American markets
    • Pricing plans (Control, Growth, Scale) with different token quotas
    • Trial period with automatic activation, invoices, payment history
    • Requests per second / minute / hour / day limits, spending control
  9. MAIL Phase 3 10 Apr 2026

    Backup mail server

    Brought up a second mail node on a separate machine. If the primary server goes down, emails aren't lost and continue to be received.

    • Automatic failover when primary is unavailable
    • Real-time mailbox synchronization between nodes
    • The email send queue withstands multi-hour outages
  10. LLM Phase 3 5 Apr 2026

    Server-Driven UI: AI draws the interface for the user

    Breakthrough thing: an AI agent can ask the gateway to show the client a ready-made working screen — a deal list, a form, statistics. Not text with a description, but a live interactive block in the right chat panel. The client sees their data, clicks buttons — the AI reacts.

    • Connected Bitrix24 — leads, statuses, bulk action buttons
    • Connected amoCRM with OAuth login and instant updates via webhooks
    • Custom rendering engine: 12+ primitives (cards, forms, tables, tabs)
    • Caching schemas for display speed
  11. LLM Phase 4 5 Apr 2026

    Widget for embedding into third-party sites

    The client can embed our AI widget on their site with a single line of code — it will display the same interactive dashboard (lead lists, action buttons). Works with any client frontend stack and won't conflict with their design.

    • A widget as a standalone web component — zero dependencies
    • Theming under the client's brand
    • Integration with custom-built sites on 1C-Bitrix
  12. MAIL Phase 2 1 Apr 2026

    Spam protection

    Connected anti-spam engine, email scoring, quarantine for suspicious messages. Outgoing campaign deliverability has grown consistently.

    • Scoring each incoming email against 50+ rules
    • Quarantine of suspicious emails with the ability to unblock them.
    • Black and white lists at the platform level and at the client level
    • Protection against common phishing schemes

March 2026

  1. LLM Phase 2 31 Mar 2026

    Messenger CRM on a separate server

    Launched the second core part of the platform — the messenger. This is where all communication channels converge, and AI agents process customer messages from any messenger as a single stream. The dashboard, dark theme, funnels — everything is in place.

    • Connected Telegram (personal account and bot), WhatsApp, Avito, and SMS
    • Sales funnels as a kanban board with drag-and-drop cards
    • Real-time notifications, push messages to Telegram bot
    • SMS bridge via mobile app — turning a regular Android into a gateway
    • Light and dark themes, single sign-on with AI gateway
  2. MAIL Phase 1 15 Mar 2026

    Basic sending and receiving of mail

    We set up a mail server: you can send emails on behalf of a client domain with signatures (so they don't end up in spam), receive incoming mail, and access mailboxes via IMAP.

    • Sending and receiving emails for any connected domain
    • DKIM digital signatures to increase trust in our emails
    • SPF and DMARC are configured automatically
    • Access to mailboxes via standard email clients
  3. TOOLS Phase 2 1 Mar 2026

    Edge nodes and storage with CDN

    We've spun up edge points — servers around the world that handle client traffic with caching. Connected an S3-compatible storage: you can host a landing page or static site via API in under a minute.

    • Edge points in Europe and Latin America
    • In-memory caching for the hottest resources
    • S3-compatible object storage
    • SDK files (widgets, scripts) are served from edge nodes

February 2026

  1. LLM Phase 1 28 Feb 2026

    Basic AI gateway: models, tools, desktop

    The very first building block of the entire platform. We stood up the core that can accept requests, select a model, calculate cost, and return a response to the client. Without this, none of the rest would exist.

    • Connected models from OpenAI, Anthropic, and DeepSeek with automatic selection based on the task.
    • We built a tool system: AI agents can now not only talk but also take action.
    • Billing counts each request in tokens and converts them to money according to the client's rate.
    • Web dashboard with chat, knowledge base, and agent settings
  2. TOOLS Phase 1 15 Feb 2026

    Our own distributed DNS

    Launched our own DNS service — no longer dependent on external providers for managing client domains. We can add zones instantly via API without contacting the registrar.

    • Four servers in different data centers for fault tolerance.
    • Full API for managing zones and records
    • Support for all record types: A, AAAA, MX, TXT, CNAME, SRV
    • Response time is consistently under 30 milliseconds.

Coming up

  1. CRM Phase 15 planned

    Rest-Block Foundation

    `BookingAppointment` can store breaks/weekends as `status='rest'`, and all appointments readers interpret them correctly — without a separate table.

    • 15-01-PLAN.md — migration of status='rest' + nullable service/client; slot-calc/counters account for rest
  2. CRM Phase 16 planned

    Booking Journal

    The operator manages a specific master's day through the "Appointments" screen — sees the timeline, creates appointments and breaks.

    • 16-01-PLAN.md — backend POST appointments + rest-blocks (admin create endpoints, tenant/IDOR guard, audit)
    • 16-02-PLAN.md — /crm/booking/{page_id}/journal route + day view UI (master select, date strip, hour timeline, rest visually separated)
    • 16-03-PLAN.md — week view + FAB sheet + 3 modals (Appointment/Break/Day off) + e2e tests
  3. CRM Phase 18 planned

    Booking i18n

    Booking templates (internal and public) render in the selected language without RU-hardcode.

  4. CRM Phase 19 planned

    booking_only Role

    There is a receptionist role with access exclusively to the booking section.

  5. CRM Phase 20 planned

    Booking Widget (embeddable)

    The CRM client embeds booking on their site with a single JS snippet; the widget adapts to the site's brand styling without us touching the code.

  6. MAIL Phase 8 planned

    Stable channel for incoming emails

    When a client replies to an email, it should reliably land back in the CRM. We're improving delivery: signing each inbound request, retrying on failures, and queuing unaccepted messages.

    • Cryptographic signature of every incoming email
    • Retries when receiver is unavailable with increasing interval
    • Queue of unsent emails for reprocessing
    • Notifications for systematic failures
  7. MAIL Phase 9 planned

    Self-managed email

    The client creates mailboxes, aliases, and forwarding rules on their own through the interface — without contacting us. Right now this is only possible via the API.

    • Web interface for creating and deleting mailboxes
    • One-click alias and redirect management
    • Setting up an auto-reply
    • View usage statistics
  8. SSO Phase 5 planned

    Migration of existing users

    Migrate all accounts from old Omra products to a single sign-on — without forcing customers to change their passwords. Transparent for the user, no disruptions in operation.

    • Import all accounts from existing services
    • Old passwords continue to work without reset
    • Gradual transition: both mechanisms work in parallel
    • Rollback plan if something goes wrong
  9. LLM Phase 12 planned

    Semantic search across customer correspondence

    All client conversations (SMS, Telegram, WhatsApp, Avito) are vector-indexed — you can search by meaning, not just keywords. AI pulls relevant pieces from history and analyzes them.

    • "Find all conversations with client Petrov"
    • "What are the most common complaints for April?"
    • "Analyze why client #123 stopped responding"
    • Retrospective indexing of all accumulated history in a single run
  10. LLM Phase 13 planned

    Return of the inspector and expense counters

    After the technical migration, two useful things in the chat got lost — visibility of the AI tools that were invoked and the token usage counter under each response. Fixing it.

    • List of tools called by AI, with parameters and results
    • Token counter and cost per each response
    • Full request history for debugging
  11. LLM Phase 14 planned

    Voice v2: CRM integration, migration to the main server

    We're scaling the voice POC to a production-grade service. AI recognizes the caller by number, manages the contact card, and the conversation is saved in CRM.

    • Search for a customer by incoming call number
    • Call recording in the customer card with transcription
    • Setting up a voice agent from the dashboard (voice, prompt, tools)
    • Stable infrastructure on the main messenger server
  12. LLM Phase 15 planned

    Cheap voice providers

    Right now voice runs on one expensive provider. We're adding alternatives — 10x savings for basic scenarios, premium voices for important clients.

    • Google's Gemini Live — ~10× cheaper for typical conversations
    • ElevenLabs — premium voices for VIP clients
    • Choosing a provider for each agent individually
  13. LLM Phase 16 planned

    AI-generated product history descriptions in 3 languages

    The release log and roadmap are currently in dry developer language. AI writes human-readable descriptions for each active day and each phase — in three languages. Ready for the public site.

    • Russian, English, Spanish
    • Description of each day with activity — what's new for the user
    • Description of each phase — what we built and why
    • Language switcher in the interface
  14. LLM Phase 17 planned

    Tariff plans v2 and full cost accounting

    Right now some models log tokens but don't convert them into money. We're building out full billing — every model in every currency, transparent pricing.

    • Cost of each model in dollars, rubles, and Argentine pesos
    • Transparent pricing on the landing page, synced with the actual database
    • Custom pricing for large clients
  15. LLM Phase 18 planned

    SMS bridge mobile app improvements

    Bugs have piled up in the Android app that handles the SMS channel — some SIM cards aren't being recognized, tokens expire and don't renew automatically.

    • Passing active SIM card numbers during device registration
    • Automatic token refresh on expiration
    • Protection against message loss during communication outages
  16. LLM Phase 19 planned

    Internal HTTPS between services

    Technical debt: services inside the platform talk to each other over an open protocol. Time to lock it down.

    • Internal proxy with TLS certificates for all backend services
    • Certificates are updated automatically
    • Compatibility with old clients during the transition period
  17. LLM Phase 20 planned

    Extended MCP server for external agents

    The MCP standard lets you connect Claude Desktop and other third-party AI agents directly to our tools. We're expanding support to full coverage.

    • Discovery of all available tools
    • Dynamic resources (real-time customer data)
    • Prompt templates for a specific agent
    • Launching sampling through our infra
  18. LLM Phase 21 planned

    Catalog of ready-made agents

    Showcase of pre-configured AI agents by industry. One-click setup, all necessary tools and knowledge base are connected automatically.

    • Sales agents for marketplaces
    • Agents for accounting and tax authorities
    • Support desk agents
    • One-click installation with automatic setup
  19. LLM Phase 22 planned

    Knowledge base search v2

    Improving search relevance across customer knowledge — hybrid strategy delivers more accurate AI responses.

    • Hybrid search: by keywords + by meaning simultaneously
    • Re-evaluating results with a second AI pass
    • Smart splitting of long documents into meaningful chunks
    • Access rights for individual documents and folders
  20. LLM Phase 23 planned

    Omra mobile app

    Native app for iOS and Android. Chat, threads, push notifications — sellers can work with customers from their phone.

    • Chat with AI and customers in the native interface
    • Push notifications about new messages
    • Managing the funnel from mobile
    • Biometric login
  21. TOOLS Phase 6 planned

    Secure HTTPS inside the platform

    Closing the last bottleneck: services must communicate via a secure protocol even within our own network.

    • TLS certificates for all internal addresses
    • The solution works at the DNS routing level and doesn't require any service rework.
    • Compatibility with old clients during the transition period
  22. TOOLS Phase 7 planned

    Migration to modern routing

    A modern traffic routing mechanism for lower latency and easier redundancy management.

    • Support for multi-path routing
    • Convenient management of failure scenarios for specific nodes
    • Reducing delays on popular routes