Everything the platform
can do

LLM agents with system prompts, personas, temperature, and reasoning effort (thinking) levels.

Create prompt versions, view history, and set the active version per agent.

Schema constrained, machine readable agent responses.

Move an agent between providers and models without rebuilding it.

Token by token delivery with live progress steps.

Token usage, cost, and usage metrics per agent.

Inspect every agent turn with token counts, trace details, and trace analysis, including tool and SQL calls with origin, arguments, and latency.

Flag anomalous conversations for human review and push them to a test suite or dataset.

Try any agent instantly, plus evaluation through test suites.

Describe a use case and generate a complete bot with agents, flows, guardrails, and knowledge suggestions.

Suggested example questions per agent.

Agents hand sub tasks to specialist agents, and pass a question on when their own answer is judged insufficient.

Continue a returning user's earlier conversation, or fetch earlier conversations, optionally summarized, as context for the agent.

Agents remember lasting facts and preferences per customer across conversations, with retention rules, categories that are never stored, an admin view and deletion per customer.

Connect to more than 300 models from OpenAI, Azure OpenAI, Anthropic Claude, Google Gemini and Gemma, Meta Llama, Mistral, Microsoft Phi, Alibaba Qwen, DeepSeek, GPT-OSS, Humain, Fanar and more.

Connect any external OpenAI compatible LLM with your own credentials.

Route model calls to the cloud region your deployment runs in for data residency, with Google models served through Vertex AI.

Automatic reasoning effort tuning and extended timeouts for deep research models.

Expensive models stay visible but disabled per plan, so you always see what is available.

Embeddings from OpenAI, Google Gemini and Qwen.

Ground answers in uploaded documents and crawled websites.

PDF, DOCX, PPTX, XLSX, CSV, JSON, Markdown, TXT, SVG, images, and Outlook .msg email files.

Crawl pages and subpages, recrawl, and review extracted content with an LLM based crawler.

Upload, preview, version control, and revert documents centrally.

Configurable chunk size and overlap with one click retraining.

Dense vector search combined with lexical BM25 and result deduplication.

PostgreSQL pgvector and file based HNSW indexes.

Register SQL storage (PostgreSQL or SQLite) that agents query through structured, typed tools.

Ingest incoming email via a dedicated tenant email address.

Answers cite the documents and pages they are grounded in, with links and source dates.

Date aware retrieval that prefers newer content when sources overlap.

Retention rules that expire and clean up outdated knowledge base content automatically.

Drag and drop BPMN style canvas with zoom, undo, and redo.

Click to append menu, copy and paste across flows and sessions, live validation.

Subflows, functions, and agents referenced from any flow.

Every publish creates a numbered version of the whole assistant with a diff against the previous one, one click rollback, and separate publishing to Test and Live, each with its own permission.

Export and import whole flows; collaborative lock against concurrent edits.

Character limits and media support flagged per channel while designing.

One flow holds localized content for every bot language.

See variables, block outputs, and usage statistics on the canvas.

Message, question (sensitive flag and DTMF), options, rating, random answer, dictionary, image or GIF, video, attachments, unexpected answer, multiple entity check, and adaptive cards.

Set variable or object, replace, calculation, counter, delay, validation, recognize intent, get history, statistics, send email, translation, language detection, redirect call, transfer call, send DTMF, schedule callback, register opt out, set lead outcome, handover, auth, PIN change, and complaint.

Equals, exists, math comparator, in set, contains, is of type, is an entity, is canonical, is similar, which channel, check user, regex, and conversation language.

Agent (LLM) block, custom function block, trigger workflow block, and reusable subflow block.

Intents, entities, utterances, patterns, and closed lists with training and publishing.

Microsoft CLU, Google Dialogflow, IBM Watson Assistant, Meta Wit.ai, and LLM based recognition.

Import intents and entities in bulk, or create intents directly from analytics data.

Measure NLU performance with automated testing and nightly retraining.

Track unrecognized utterances and unexpected answers for continuous improvement.

Agentic workflows with an agent loop of up to 25 steps that calls tools on its own and changes its plan based on what each tool returns.

Instructions, configuration, validation check, and activate or deactivate toggle.

Run analytics, a full audit trail per run, and downloadable run data.

Deferred, condition triggered tasks (do X when Y) with scheduled rechecks, expiry, and dry run mode.

Per tool approval: an agent runs the tool automatically, asks the customer to confirm, or waits for an employee to approve, optionally only above an amount. Every approval, rejection and timeout is logged.

Set per tool whether an agent may run it automatically, needs approval first, or may not use it at all.

Start workflows from dialog flows, on a schedule, from a webhook, an incoming email or a new file, or via API tokens.

Workflows keep their own state between runs, such as the cases already handled, and agentic tasks track a goal for one customer across sessions.

Web search (Exa, Google), Wikipedia, web browsing, image generation (DALL-E), image search, file generation (PDF, Word, Excel), RAG retrieval, SQL querying, payment links, lead capture, handover, and call control for voice agents (end call, transfer, send DTMF, schedule callback, register opt out, and record lead outcome).

Connect external MCP servers (SSE or HTTP) and expose their tools to agents, with defaults for HubSpot and Notion.

Expose your SQL storage as a hosted MCP server with typed query tools, protected by an API key and rate limiting.

Publish Blits.ai agents with an Agent Card so agents on other platforms can hand them tasks, and call external A2A agents, for example on Salesforce or ServiceNow, as tools.

Value mapping, taxonomy translation, and place name recognition improve answers over SQL data.

Control which rows and records an agent may use, in soft or strict mode.

HubSpot, Freshdesk, Zoho, Notion, Moneybird, Odoo, Dynamics 365, Office 365, OneDrive, Power BI, SharePoint, and Teams.

Salesforce, SAP, ServiceNow, Workday, Zendesk, Oracle, Jira, Snowflake, Databricks, UiPath, DocuSign, Okta, Slack, Google Workspace, and many more.

Per system OAuth for Notion, HubSpot, Zoho, and generic providers.

REST, SQL, and custom code in an isolated JavaScript sandbox, with versioning, publishing, and global functions.

Payment provider webhooks, incoming email webhook, and ElevenLabs webhook.

Drop the web widget into any WordPress site.

Embeddable, snippet based chat widget for any website.

Two way messaging via the Meta Cloud API and via Twilio.

Text conversations over Twilio SMS.

Twilio Voice with real time streaming STT and TTS, OpenAI Realtime pipeline, call transfer, and DTMF input.

Proactive AI phone calls at campaign scale, through Twilio or your own SIP telephony, with compliance and lead capture built in.

Teams channel with automatic bot provisioning and reliable error handling.

Direct integrations for Slack, Telegram, and Facebook Messenger.

Embed the assistant as a native in app experience, with its own channel analytics, host context variables, and offer redemption flows.

Inbound and outbound email conversations.

Voice skills for Amazon Alexa.

Photorealistic 3D avatar channel streamed to the browser.

Programmatic channels to drive conversations from any application.

In platform preview and a shareable external link, switchable between bubble, mobile, WhatsApp, and digital human.

Salesforce, Freshdesk, and Zoho SalesIQ as conversation channels.

AI agents place outbound phone calls, open with a spoken disclosure, run the full conversation, and capture the outcome.

Upload a contact list, set the calling window, and launch a campaign that dials, paces, and retries automatically.

Call through Twilio, or connect your own telephony over SIP (Cisco and any SIP capable estate) so no third party sits in your call path.

Detect voicemail and apply the campaign policy: hang up, or leave an identified message with a callback number.

Hand a warm lead to a configured agent or team, or register it through the existing handover integrations.

The contact picks a better moment and the campaign calls back inside the allowed window.

Qualify each lead against a schema, capture structured lead data, and export results as CSV.

Spoken AI disclosure on every call, instant opt out and suppression, statutory calling windows, attempt caps, and a per attempt audit trail.

Per campaign call recording with a recording announcement and configurable retention.

Per campaign spend limits with automatic pause, plus an instant global stop for every campaign.

Typing indicator and live progress steps.

Automatic voice detection, start and stop, and push to talk microphone modes.

Streaming audio playback with an audio toggle.

Optionally hide voice transcripts in the chat: voice turns show a voice message placeholder with a waveform instead.

Markdown, HTML, quick replies, images, carousels, adaptive cards, code blocks, and reasoning or tool call elements.

More than 45 card kinds: flights, hotels, food, retail, orders and receipts, loyalty points, offers with redemption, credit cards and card benefits, telco usage, balances, charts, and product recommendations, each with per card prompt instructions and translated labels.

Connect a user data provider and verified balances, loyalty points, and tiers become variables, badges, and cards in the conversation.

Attach and upload files in the conversation.

Thumbs up and down with optional comments, configurable per channel.

Colors, avatar, icon, title, logo, custom CSS, prebuilt themes, position, size, and a Powered by toggle.

Teaser messages with a notification badge.

Progress updates in the language of the user, with detailed, simple, or off modes.

Configurable acknowledgment and follow up messages per language while the agent works.

Language switcher, per language placeholders, and example prompts.

Resume after page refresh, starter prompts, and idle timeout handling.

Optional hashing of external user IDs, shadow DOM isolation, and CSP safe embedding.

Azure, Google, Amazon Polly, IBM Watson, OpenAI, ElevenLabs, Deepgram, Resemble, Murf, Cartesia, Mistral Voxtral, Silma, Gradium, and Gemini, plus your own custom endpoint, with custom voices and fine controls.

OpenAI, Deepgram, ElevenLabs, Google and Gemini, Azure, Speechmatics, Gradium, a self hosted option, and your own custom endpoint.

Agent emotions map to emotion capable voices.

Fast, cost efficient repeated audio with cache management.

Generate and test speech from text in the platform.

Numbers, dates, and amounts rendered in the caller's dialect before synthesis, including Arabic regional dialects.

Speech input rewritten in the background by a lightweight model for cleaner transcripts without extra latency.

Connect any OpenAI compatible TTS or STT endpoint with your own key, including self hosted models.

Speech recognition provider and dialect configured per bot language.

Silence detection calibrates to the background noise, so soft speakers are not cut off.

Users can interrupt the bot while it is speaking and the bot stops talking to listen.

WhisperX on GPU: word level timestamps, language detection, and who spoke when, on Blits.ai infrastructure.

Photorealistic avatars via Unreal Engine Pixel Streaming with lip synced speech.

Agent emotions drive avatar animation and expressive voice.

Multiple avatar personas, each on its own GPU backed instance.

WebRTC streaming that works through firewalls and NAT.

Agents send payment links and complete transactions in the chat flow.

Stripe, Mollie, Adyen, and Qi, selectable per tenant.

Pay by link with success and cancel URLs and expiry.

Hold and later capture, the approval primitive for agent initiated purchases.

Card data is vaulted at the PSP and never stored on the platform.

Signature verified webhooks keep payment status in sync, with multi currency support.

Isolated payment service, tokenization at the gateway, encrypted credentials, audit logging, amount ceilings, idempotency, and velocity limits.

LLM judged content checks with admin authored policies plus a sensible platform default.

Curated lexicons for hate speech, self harm and violence, and profanity detection across 25 languages with evasion resistant matching.

Regex rules, quality checks, and output sanitization.

Deterministic enforcement of blocked topics, independent of model judgment.

Automatic masking at the gateway with per bot patterns and encryption or redaction modes.

Credit card detection and tokenization in free text chat.

Protect public endpoints against floods and abuse.

Automated conversation evaluations against agents, bots, and workflows through the real chat pipeline.

Conversation sets run as one continuous session, with per turn intent notes for the grader.

Deterministic rules and LLM grading with a configurable prompt, selectable model, reasoning, and KB evidence.

Pass rate, per turn verdicts, grader reasoning, follow up exchanges, run history, and regrading.

Import from JSON with a downloadable template.

Scheduled health checks with deterministic or LLM expectations and email or webhook alerts on failure and recovery.

Exercise the live widget under simulated network and CPU conditions.

Upload audio and compare speech to text providers on word error rate.

Test tool discovery health for external MCP servers.

When a model provider fails, the turn moves to a model from another provider. Operators can switch this off per environment.

Rate limit and quota errors are retried with increasing waits, and model calls time out instead of hanging.

A failed tool call goes back to the agent, which tries other arguments, another tool or another route.

Lost MCP connections reconnect to the same server automatically and the call is retried, with a 15 second timeout per call.

Answers that score below the quality threshold are rewritten before the customer sees them.

Agents that plan to answer without consulting the knowledge base are sent back to plan again with the knowledge search, or retrieval always runs first.

If a turn still fails, the customer gets a configured message instead of silence.

Promote an assistant to production only when a selected test suite, including latency checks, passes on the new version.

99.5% monthly availability for cloud hosted deployments, measured and published on status.blits.ai.

Liveness and readiness checks on every service, including database connectivity.

KPI tiles for interactions, recognition, unique users, satisfaction, and sentiment versus the previous period.

By channel, top intents, recognition per engine, NLU performance, untrained questions, and system performance.

Turn untrained questions and unexpected answers into new intents from the dashboard.

Calculated statistics, custom dashboard widgets, time ranges, and CSV export.

Transcript viewer with export controls and permission gating.

Watch live conversations and take over in real time, voice enabled on telephony.

Thumbs and comments with summaries, filtering, screenshots, and captured call audio (the caller's voice and the bot's spoken reply), plus conversation level CSAT.

Flow statistics, LLM usage statistics, and per agent metrics.

Interactions measured per tenant, bot, and domain, with an AI cost calculator and interaction calculator in the platform tools.

Via Amazon, Google, IBM, and Microsoft engines.

From a flow block, an agent tool, or automatic rules.

Take over from the admin console, including on voice calls.

A human agent joins the conversation in the web chat widget, with agent messages clearly marked.

Salesforce (chat and SOAP), Freshdesk, and Zoho SalesIQ.

Configurable per bot with availability checks.

Per bot language list with per language localization of all content.

Detect language and switch mid conversation, without losing the conversation.

Via Amazon, Google, IBM, and Microsoft.

Support for regional dialects and languages, including normalization and region specific voices and models.

Every record and query is scoped to the tenant.

Consent gating, privacy policy link, user data removal, retention controls, and sensitive field flags.

Mask user messages and AI data with configurable, testable patterns.

Hash user IDs so people are identified by a pseudonym instead of personal data.

PCI DSS Level 1 certified payment architecture.

User audit logs, per task agentic audit trails, and tamper evident event logging.

JWT with short lived refresh tokens, brute force protection, email confirmation, and password reset.

Microsoft (Azure AD / Entra) and Google SSO.

Multi factor authentication on admin login.

Keys hashed and validated server side, with signed service to service authentication.

A timeline per conversation of routing decisions, tool calls with masked inputs and results, guardrail checks, handovers and the model used, exportable as JSON or streamed over OpenTelemetry to your SIEM.

Master tenant and subtenant structure; users span multiple tenants and switch between them.

60+ fine grained permissions, custom roles per tenant, per bot roles, and view versus manage tiers.

Separate publish permissions for Test and Production.

Users, roles with a permission matrix, bots, API keys, and subscription.

Every settings change is recorded: who changed it, what changed, and when.

Drive conversations from any application, with initialize, message, liveness, erase, and analysis endpoints.

Tokens for agents, bots, and workflows, each with its own type and endpoint.

Public API for tenant interaction counts.

Run agents programmatically from your own applications.

Kubernetes native deployment to any public or private cloud, with managed production regions in the EU, Qatar, Singapore, the UAE, the US and Australia.

Local deployment via Docker Compose (MongoDB, PostgreSQL with pgvector).

Routing, streaming proxy, PII masking, card tokenization, and rate limiting in front of the platform.

Nightly NLU retraining, tenant lifecycle checks, and plan enforcement.

Graceful shutdown drains running conversations before an instance stops.

Curated use cases by sector with capabilities and supported channels.

Encrypted bot export and import, plus intents, entities, and functions.

Google Cloud project setup for Google based channels.

Produce audio from text on demand.

Built in guide and collapsible navigation with tenant switching.