> ## Documentation Index
> Fetch the complete documentation index at: /docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Product updates

> Stay updated with the latest BoostGPT features and improvements

**November 2026**

## Agents are now AI teammates

The dashboard, emails and plans now say **AI teammates** instead of agents, and the people in your workspace are
**members**. Nothing moved: your AI teammates, Sites and links work as before.

The API and SDKs keep the names `bot` and `agent`, so your code doesn't need to change.

**October 2026**

## Every AI call metered, three models retired

**Credits.** Plans now state their allowance as credits per month. Chat titles, page suggestions, tool
descriptions for imported APIs and replies you stop mid-stream now use credits like every other AI call, in line
with their tokens. Calls made with your own API keys, and models on your own Ollama host, still use none.

**Retired:** `claude-sonnet-4-5` (Anthropic retires it on November 30, 2026), `gemini-3-flash-preview`,
`kimi-k2.5` and `gemini-3.1-flash-lite` (Google shuts it down on May 7, 2027). Agents that used them were moved to
`claude-sonnet-5-5`, `gemini-3.6-flash`, `kimi-k2.6` and `gemini-3.5-flash-lite`.

**Scheduled prices.** Gemini 3.8, 3.7 and 3.6 Flash double in price on January 1, 2027, and their credits follow
on that day. An Ollama model now needs your Ollama host saved in Integrations.

**October 2026**

## New models, and credits from real prices

**Added:** `gpt-6.1-sol` (near-Astra performance at Sol's price) and `claude-sonnet-5-5` (Anthropic's new
speed-and-intelligence model), both 8 credits.

**Retired:** `gpt-5.1` and `gpt-5.4-nano`, which OpenAI shuts down on April 1, 2027. Agents that used them
were moved to `gpt-6-sol` and `gpt-6-luna`.

**Credits recalculated.** Every model's credits now come from its provider's published price for a typical
agent step — about 31K prompt tokens, most of them cached, and 465 output tokens. Most models cost more
credits than before, some cost fewer, and GPT-6 Luna is unchanged at 1. See
[Model Comparison](/docs/providers/model-comparison.md) for every figure.

**September 2026**

## Calling as a service

`chat()` accepts `purpose: 'service'` for calls made by **your own backend** on behalf of many end
users — a WhatsApp assistant, a helpdesk, a community bot — rather than by one person chatting.

A service call may pin a workspace with `workspace_uuid` and keep a `chat_mode` such as `'edit'`,
so your agent can keep durable notes across conversations and read them back later. A pinned
workspace is remembered against the `chat_id`, so a turn that pauses for tool execution resumes in
the same workspace — there is nothing extra to send to `executeTool()`.

It is deliberately **not** a builder conversation: it never receives the Site builder persona, and
never gets `read_conversations`, which reads what other people have asked. Workspace scope, never
cross-conversation reach.

`purpose` is requested, not asserted — it is granted only when your API key belongs to someone who
runs the project. It is also opt-in: **existing integrations are unaffected**, because a call that
sends no `purpose` behaves exactly as before.

Service workspaces have their own plan allowance, counted separately from the workspaces you create
in the dashboard and shown in your account usage.

**August 2026**

## CRM, calendar booking, and follow-ups

The CRM suite has been retired:

* **CRM workspace** — contact management, deals, Kanban pipelines, stage editing, notes, activity history and custom intake fields, along with their screens and APIs.
* **Tasks, reminders and follow-ups** — follow-up tasks with due dates, priorities, reminder status, snoozing and backlog views.
* **Calendar booking cards** — the `calendar_check_availability`, `calendar_request_booking` and `calendar_create_event` tools and the in-chat booking card they rendered.

Three things people confuse with these, which all still work: the **Google
Calendar connector** is untouched, so agents can still read and write a calendar
through it. **Workflow automations** remain — minus their CRM triggers and CRM
action blocks. And **lead capture** is unaffected: Site form submissions still
collect signups, visible under **Leads** in the Site builder.

## Workspaces and the public agent

The Sites builder now owns everything that needs a workspace, so the workspace
surfaces that predate it have been retired from the public agent:

* **Workspace scope** — the per-account / per-chat setting is gone, along with the `workspace_scope` field on agent create and update. **Every conversation now gets its own isolated workspace**, authenticated or not. Agents previously set to per-account keep their existing workspace records, but new conversations no longer resolve to them. This supersedes the March 2026 Workspace Scope Toggle.
* **Workspaces on the public agent page** — the file panel, the workspace switcher and the composer workspace toggle are now visible to project members only. Visitors chatting with a shared agent, a widget or an embed never see workspace UI, and their conversations no longer resolve or create one.
* **Chat modes on the public agent** — public conversations are always **Chat**. The Chat/Edit/Plan toggle is limited to project members, the widget no longer receives `chat_mode` in its config and no longer renders a mode toggle, and `PUT /v1/bot/chat/mode` rejects `edit` and `plan` on conversations that are not the Site builder's. `chat_mode` on an agent is now a default for your team only, not for visitors.
* **Publishing and custom domains in the workspace file panel** — removed. Custom domains now live in the Site builder under **Settings → Publishing**, where they are plan-gated with an upgrade prompt instead of a raw error.

Two notes on settings you will still see. **Show workspace files** no longer
affects the public agent page — it is retained for compatibility and no longer
changes what visitors see. **Default chat mode** now applies to your team's
conversations only.

Agents that are not Sites are unaffected apart from workspace scope: their
members keep full agent, edit and plan modes, including in the inbox.

## Workspace search tools

`grep` gained the flags a shell user expects: `files_without_match` (which files LACK a
pattern), `invert_match`, `multiline` for patterns spanning lines, `case_sensitive`,
`context` for surrounding lines, and glob `include`/`exclude` in place of the old
extension-only filter. A per-file match cap that used to truncate silently now reports
`truncated: true`.

**`glob` is new** — find files by path pattern, e.g. `*.html` for every page. `ls` gained
`recursive`.

**`command` is now `bash`.** Same sandbox, same behaviour, same approval rules; the name
simply says what it is. Existing conversations that recorded `command` still render, and
the `command` parameter it takes is unchanged.

## Agent collaboration

Agent-to-agent messaging has been retired:

* **`message_agent` tool** — the built-in tool that let one agent queue an asynchronous message to another agent in the same project.
* **Agent collaboration setting** — the toggle under agent settings, and the `inter_agent_communication` field on the agent API.

Nothing else changes. **Scheduled tasks** and **notifications** are untouched, and
agents still run on a schedule and notify people as before — only the agent-to-agent
delivery path is gone.

If you set `inter_agent_communication` when creating or updating an agent, the field is
now ignored rather than rejected, so existing calls keep working.

**August 2026**

## Heartbeats are now scheduled tasks

Heartbeats and scheduled tasks were the same engine written twice — same helpers, same queue, same
log table — and one bug fix landed in only one of them. They have merged.

> [!WARNING]
>
> `bot/heartbeat/*` no longer exists. The endpoints below take its place, and `heartbeat_id`
> is now `task_id`.

A heartbeat WAS a task with a cron, so the capability survives; only the name changed. Nine SDK
methods are renamed, and each keeps its shape — porting is a find-and-replace:

| 5.x | 6.0 |
| - | - |
| `createHeartbeat` | `createTask` |
| `fetchHeartbeat` | `fetchTask` |
| `fetchHeartbeats` | `fetchTasks` |
| `enableHeartbeat` | `enableTask` |
| `disableHeartbeat` | `disableTask` |
| `triggerHeartbeat` | `triggerTask` |
| `deleteHeartbeat` | `deleteTask` |
| `fetchHeartbeatLogs` | `fetchTaskLogs` |
| `fetchHeartbeatStats` | `fetchTaskStats` |

* **`cron_pattern` still works** on `createTask` and still means a recurring schedule — it is
  translated to `schedule_type: 'cron'` for you. `schedule_type` and `schedule_config` are
  available directly if you want an interval, a daily or weekly time, or a one-shot.
* **`createTask` takes a `name`.** Omit it and the first 60 characters of the prompt are used, so a
  ported `createHeartbeat` call still works.
* **`credit_budget` is gone.** It capped a subscriber's spend on a heartbeat, and subscriber-owned
  tasks were retired — an owner-created task was never charged by it in the first place.

Tasks appear on the agent's task board alongside everything else it runs, so a schedule that used
to be invisible outside the API is now something you can see and change.

**July 2026**

## Removed features

Several legacy surfaces have been retired now that the Sites builder covers their use cases:

* **Agent Interface** — the `set_interface` tool, chat-level and bot-level interface files, and the "Set as Chat Interface" / "Set as Default Interface" workspace actions. Build a Site instead.
* **Bridge SDK** — the auto-injected `window.BoostGPT` JavaScript SDK and its `chat`/`workspace`/`tools`/`schedule`/`bot`/`ui` namespaces.
* **Interface templates** — publishing workspace HTML files as shareable interface templates, and the subscriber template picker.
* **Dashboard assistant platform tools** — the 17 platform tools and the dashboard assistant they powered.
* **`search_workforce`** — the marketplace search tool and its `workforce_tools` gate.
* **Widget sidepanel open style** — along with the page-push/docking behavior, presentation mode, and agent page guidance (`page_control` / `page_context`) that only ran in sidepanel embeds.
* **Event Triggers** — the `boostgpt.trigger()` API, the standalone `trigger.js` script, the `POST /v1/bot/trigger` endpoint, the Events screen, trigger analytics, and the per-plan trigger quota. Use the always-on widget and greeting messages instead.
* **Agent marketplace** — the `/listings` and `/admin/listings` APIs, listing ratings, the public marketplace browse page, and the global `users.is_staff` flag. Project member roles are unaffected.

The widget `open_style` options are now `normal` and `dock`. Widgets configured as `sidepanel` fall back to `normal`.

**May 2026**

## Calendar booking cards

> [!NOTE]
>
> Removed in August 2026. The `calendar_*` tools and the in-chat booking card have been retired. The Google Calendar connector is unaffected.

Agents can now complete booking flows inside chat:

* **Availability checks** - `calendar_check_availability` returns real open slots for a duration, date range, and timezone
* **Interactive booking card** - `calendar_request_booking` renders selectable slots and waits for the visitor to choose a time
* **Calendar event creation** - `calendar_create_event` creates the event after a slot is selected
* **Confirmation email** - attendees receive a booking email with an `.ics` file and calendar links when an email address is available
* **Timezone-safe display** - event responses include `display_time` so final answers do not recalculate UTC timestamps incorrectly

The booking card is treated as user input, not a tool approval step, so public visitors can choose a slot naturally.

## Deploy channel states

The Deploy page now focuses on launch readiness and channel state:

* **Launch readiness** - hosted page, widget, CRM, calendar, workflow, workspace, and connector checks
* **Channel states** - enabled, setup required, plan-gated, and coming soon
* **Agent email state** - address, auto-reply, and notification status
* **Workspace publishing** - published workspace URLs and custom domain status are surfaced alongside other deployment channels
* **Router SDK channels** - Slack, Discord, Telegram, WhatsApp, and Crisp are grouped as advanced deploy channels

## Dashboard assistant platform tools

> [!NOTE]
>
> Removed in July 2026. These platform tools and the dashboard assistant are no longer available.

The dashboard assistant can now use internal platform tools to inspect and configure agents:

* list manageable agents and projects
* check setup readiness
* check plan limits
* update agent feature switches
* apply CRM and workflow templates
* configure calendar, widget, email, connectors, and workspace publishing
* prepare test conversations and export sanitized `.boostgpt` templates

Tools that change user-facing behavior require confirmation before they run.

See [Deploy](/docs/mira/channels.md) and [Built-in Tools](/docs/sdk/core/built-in-tools.md) for details.

**May 2026**

## Workflow automations

> [!NOTE]
>
> Updated in August 2026. Workflow automations still run, but the CRM triggers and the CRM action blocks described below were removed with the CRM.

Agents now include bot-level workflow automation for turning lead and subscriber events into follow-up:

* **Business workflow editor** — build trigger -> condition -> action workflows from the Automations tab
* **Event triggers** — run workflows from CRM contact/deal/task events, subscriber lifecycle events, billing events, and incoming webhooks
* **Scheduled workflows** — run workflows on a recurring minute interval
* **Action blocks** — notify owners, send webhooks, create/update CRM contacts, deals, tasks, notes, and run connector tools
* **Template mapping** — map event payload data into actions using `{{field}}`, `{{payload.field}}`, and `{{steps.step_id.result}}`
* **Run logs** — inspect recent workflow runs, status, failures, and execution output

Workflow-generated CRM actions suppress recursive workflow triggers so accidental loops do not spiral.

## CRM workspace

> [!NOTE]
>
> Removed in August 2026. Contacts, deals, pipelines, tasks and reminders have been retired. Lead capture continues through Site forms.

BoostGPT agents now include a lightweight CRM for turning conversations into managed leads and sales follow-up:

* **Lead inbox** — capture contacts from agent conversations with customizable intake fields
* **Contact management** — edit, archive, search, and review contact details, notes, custom fields, and activity
* **Deals and pipelines** — create deals, move them across Kanban stages, edit stage names/status, reorder stages, and delete deals or stages safely
* **Tasks and reminders** — create follow-up tasks with due dates, priorities, reminder status, snoozing, and background reminder delivery
* **CRM webhooks** — subscribe to CRM contact, deal, pipeline, custom field, task, note, and reminder events from the Webhooks integration
* **Reporting** — view basic CRM totals for contacts, deals, pipeline value, and tasks
* **Agent tools** — agents can create/update CRM contacts, notes, deals, and follow-up tasks during conversations

CRM tools are plan-gated and can be enabled from agent settings.

## Tool access controls

Connector and built-in tools now support role-aware access:

* **Public-safe tools** — lead capture, booking, workspace/sandbox tools, scoped scheduling, and scoped notifications can run in public agent chats
* **Staff-only tools** — private CRM search, deal lists, reports, task lists, and any connector tool marked `staff_only` are limited to owners, admins, and moderators
* **Scoped CRM writes** — public users can only update CRM records connected to their current chat
* **Safer notifications** — public users cannot force arbitrary notification recipients
* **Custom connector metadata** — connector builders can mark tools with `access`, `tool_access`, `boostgptAccess`, or `boostgpt_access`

See [Tool Access](/docs/integrations/custom/tool-access.md) for custom connector examples.

**April 2026**

## Agent templates

Export any agent as a portable `.boostgpt` file and import it to create a new agent or update an existing one:

* **Full config export** — captures instruction, model, capabilities (memory, sandbox, heartbeat, etc.), branding, tool connections, artifact settings, and upload rules
* **One-click import** — import when creating a new agent or apply to an existing agent from Settings → Agent Template
* **Predefined connector auto-provisioning** — tools like HubSpot, Slack, and Twitter are automatically wired up during import; connect your credentials in the Tools tab
* **Credential-safe** — API keys and OAuth tokens are never included in exported files
* **Plain JSON** — `.boostgpt` files are human-readable and can be version-controlled alongside your code

Export from **Settings → Agent Template → Export template**. See the [Agent Templates guide](/docs/mira/templates.md) for the full format reference.

**April 2026**

## Chat widget

Embed any agent on any website as a floating chat bubble — one script tag:

```html theme={null}
<script src="https://embed.boostgpt.co/widget.js" data-bot-id="YOUR_BOT_UUID"></script>
```

* **Full agent experience** — workspace, tools, file uploads, and custom interfaces all work inside the widget
* **Branding** — bubble inherits your accent color, avatar, and greeting messages. 16 built-in SVG icons or use your bot avatar
* **Greeting bubbles** — stacked message cards from your branding settings appear above the floating button
* **New chat button** — accent-colored header with bot name, reload icon, and close button
* **Chat persistence** — conversations persist across page refreshes via localStorage
* **Mobile responsive** — full-screen overlay on small viewports
* **Auto-open** — optionally open the widget after a configurable delay
* **Programmatic API** — `open()`, `close()`, `toggle()`, `hide()`, `show()`, `destroy()`, and event listeners
* **Framework support** — setup guides for HTML, Next.js, Nuxt.js, React, Vue, and SvelteKit

Configure the widget in **Branding → Design → Chat Widget**. Preview it live in the dashboard with the Page/Widget toggle.

See the [Chat Widget documentation](/docs/site/chat/installation.md) for setup and the [API reference](/docs/developers/widget-api.md) for programmatic controls.

**April 2026**

## Widget SSO

Auto-identify users in the chat widget without requiring them to log in again:

* **JWT-based** — your backend signs a token with the bot's SSO secret. The widget verifies it server-side.
* **Auto-provisioning** — if the user doesn't exist, BoostGPT creates their account and subscriber record automatically
* **Chat history** — identified users get persistent conversations and user memory across sessions
* **Skip login gates** — SSO users bypass `require_login` settings
* **Secure** — tokens expire (1h), are signed (HS256), and the secret is never exposed client-side

Generate your SSO secret in **Deploy → SSO Secret**.

**March 2026**

## Agent marketplace

> [!NOTE]
>
> Removed in July 2026. The agent marketplace and its listing APIs have been retired.

Bot owners can now submit their agents to a public marketplace:

* **App-store model** — agents go through `draft → pending_review → approved` status flow with staff review
* **Ratings & reviews** — users rate agents 1–5 stars with optional written reviews. Cached aggregates for fast display.
* **Public browse** — search by name, category, or featured status. Sorted by rating and views.
* **Creator dashboard** — view all your listings, update metadata, resubmit rejected listings, or withdraw
* **Admin review** — staff can approve, reject (with notes), or suspend listings

**March 2026**

## Pluggable agent interfaces

> [!NOTE]
>
> Removed in July 2026. Agent interfaces and the Bridge SDK have been retired — use the Sites builder instead.

Agents can now build custom HTML interfaces and activate them as the chat UI — replacing the default message view with a fully interactive app:

* **`set_interface` tool** — agents call this built-in tool to activate any workspace HTML file as the chat interface
* **Bridge SDK** — auto-injected JavaScript SDK (`window.BoostGPT`) that lets custom interfaces communicate with the chat system via postMessage
* **Bidirectional API** — interfaces can send messages (`chat.sendMessage`), read files (`workspace.readFile`), listen for responses (`chat.onResponse`), approve tools, and more
* **Per-chat or default** — set an interface for a single conversation or as the bot's default interface for all new chats
* **Real-time updates** — `workspace.onFileChange()` auto-refreshes the interface when the agent updates data files
* **Dashboard controls** — "Set as Chat Interface" and "Set as Default Interface" context menu actions in the workspace file explorer

**Bridge SDK namespaces:** `chat`, `workspace`, `tools`, `schedule`, `bot`, `ui` — full programmatic access to the agent's capabilities from within custom HTML.

This enables agents that deliver full applications — fitness trackers, dashboards, interactive tutorials, data visualizations — not just text responses.

**March 2026**

## Workspace scope

> [!NOTE]
>
> Removed in August 2026. The setting is gone and every conversation now gets its own isolated workspace.

Bot owners can now choose how workspaces are scoped for authenticated users:

* **Per Account** (default) — one workspace shared across all conversations. Files persist between chats.
* **Per Chat** — each conversation gets its own isolated workspace. Ideal for tutoring, code reviews, and sandbox experiments.

Anonymous users always get per-chat workspaces regardless of this setting.

Configure in **Settings → Workspace Scope**.

**February 2026**

## Code sandbox

Agents can now run code in isolated [Microsandbox](https://github.com/nicholasgasior/microsandbox) microVMs — not simulated, actually executed:

* **`command` built-in tool** — agents write and execute shell commands, Python scripts, or Node.js code in a secure sandbox
* **Isolated microVMs** — each execution runs in its own container with 512MB–2GB RAM, 1–2 CPUs, and 30–180s timeout (varies by plan)
* **File sync** — workspace files are pushed to the sandbox before execution, results synced back
* **Runtime auto-detection** — automatically selects Python or Node.js based on file extensions in the workspace
* **Output capture** — stdout, stderr, and exit codes returned to the agent. Non-zero exit codes flagged with warnings.
* **Large file support** — files over 60KB are chunked to avoid system limits

Enable in **Settings → Code Sandbox**. Requires the Starter plan or above.

**February 2026**

## Agent email

Every agent now gets its own email address:

* **Unique address** — each agent receives an email like `youragent@boostgpt.page`
* **AI-powered replies** — users email your agent and receive AI responses based on the agent's instruction and training data
* **Dashboard integration** — email conversations appear alongside chat conversations in your dashboard
* **Subscriber capture** — email senders are automatically added as subscribers
* **Custom email templates** — customize verification, welcome, password reset, subscription, and credit notification emails with your branding

Configure email templates in **Branding → Emails**.

**February 2026**

## Built-in agent tools

> [!NOTE]
>
> Updated in August 2026. `message_agent` has been removed; the other tools listed here still exist.

Agents in `agent` reasoning mode now have access to **22 built-in tools** across 8 categories — no configuration required:

* **Knowledge** — `search_memory` for vector search through training data, `screenshot` for capturing web pages, `web_fetch` for extracting page content as markdown
* **User Memory** — `remember`, `recall`, `list_memories`, `forget` for managing per-user long-term memory
* **Files** — `write`, `read`, `edit`, `grep`, `delete` for sandbox workspace operations
* **Planning** — `create_plan`, `update_plan`, `read_plan`, `accept_plan` for structured multi-step workflows
* **Communication** — `ask_user` for structured questions with selectable options, `notify_user` for proactive notifications, `message_agent` for cross-agent messaging
* **Execution** — `command` for running code and shell commands in an isolated sandbox
* **Automation** — `schedule_task` for recurring scheduled tasks
* **Utility** — `think` for internal reasoning without user-visible output

Tools use three approval levels: **Auto** (executes immediately), **Conditional** (configurable per agent), and **Always** (requires explicit user approval).

See the [creator guide](/docs/mira/built-in-tools.md) for an overview or the [SDK reference](/docs/sdk/core/built-in-tools.md) for detailed parameter schemas.

**February 2026**

## Workspaces

> [!NOTE]
>
> Updated in August 2026. Isolation is per-conversation for everyone — the per-user scoping described below no longer applies — and workspaces are visible to project members only.

Agents now have a sandbox file system — isolated workspaces where they can create, read, edit, search, and delete files:

* **Built-in tools** — `write`, `read`, `edit`, `grep`, and `delete` tools available to agents in `agent` reasoning mode
* **Per-user isolation** — authenticated users get their own workspace per agent, anonymous users get a chat-scoped workspace
* **File explorer** — built-in file tree UI with syntax-highlighted code viewer
* **Real-time sync** — files created or edited by the agent appear instantly in the workspace explorer
* **REST API** — full workspace and file management via the API for programmatic access

See the [Workspaces documentation](/docs/sdk/core/workspaces.md) for usage details.

**February 2026**

## Agent reasoning mode

The reasoning system has been simplified to three modes:

* **`auto`** — automatically selects the best approach for each request
* **`standard`** — quick, single-pass answers (1x credit)
* **`agent`** — autonomous multi-step reasoning with tool use (up to 10x credit)

The `agent` mode replaces the previous `interactive`, `stepwise`, and `react` modes with a unified autonomous agent loop that can read files, search, edit code, execute tools, and plan multi-step workflows.

**Deprecated:** `interactive`, `stepwise`, and `react` reasoning modes are no longer supported. Use `agent` instead.

**February 2026**

## Plan chat mode

Conversational "plan-then-execute" workflow for multi-step tasks:

* **Plan mode** (`chat_mode: 'plan'`) — AI proposes a step-by-step plan before executing any tools
* **Approve / Modify / New** — approve to execute, reply with changes to revise, or start a fresh plan
* **Tool-aware plans** — the AI considers all available tools (built-in and connected integrations) when generating plans
* **Direct answers** — if no tools are needed, bypasses planning and responds directly

See the [Chat API documentation](/docs/sdk/core/chat-api.md#plan-chat-mode) for usage details.

**February 2026**

## Edit chat mode

Agent-powered code editing through the agent reasoning pipeline:

* **Edit mode** (`chat_mode: 'edit'`) — routes your request through the agent, which uses the built-in `edit` tool for targeted file changes
* **Toggle switch** — Chat/Edit/Plan toggle next to the send button
* **Edit button** — per-message "Edit" action on system messages with code blocks
* **Smart editing** — the agent analyzes your request, locates the right file, and applies precise string replacements
* **Multi-step edits** — the agent can chain multiple edits across files in a single turn

See the [Chat API documentation](/docs/sdk/core/chat-api.md#edit-chat-mode) for usage details.

**February 2026**

## Heartbeat scheduling

> [!NOTE]
>
> Superseded in August 2026. Heartbeats and scheduled tasks have **merged** — `bot/heartbeat/*`
> no longer exists and `heartbeat_id` is now `task_id`. Everything described below still works; it is
> a [scheduled task](/docs/sdk/core/tasks.md) now. (An earlier note here said heartbeats were unaffected by
> the `message_agent` removal, which was true of that change and is no longer true of this one.)

Transform agents from passive chatbots into proactive autonomous assistants:

* **Autonomous execution** — agents wake up on a timer, execute a prompt, use connected tools, and report back
* **Flexible scheduling** — configurable intervals, active hours, active days, and timezone support
* **notify\_user** — agents can send notifications (email, in-app) to subscribers during execution
* **schedule\_task** — agents can create their own recurring tasks for follow-up actions
* **message\_agent** — agent-to-agent async communication for multi-agent workflows

## Safety controls

* **Circuit breaker** — auto-pause after consecutive failures (configurable threshold)
* **Credit budgets** — per-heartbeat spending limits that reset monthly
* **Plan-tier caps** — maximum heartbeats per agent and minimum intervals enforced per plan
* **Bot-level toggle** — per-agent heartbeat enable/disable in Settings → Features & Options

## Execution dashboard

* **Logs** — every execution with status, duration, tokens, and credits charged
* **Activity chart** — daily visualization of execution volume
* **Stats** — success rate, type breakdowns, and time-series data

## API access

Full heartbeat management via the external API (`/v1/bot/heartbeat/*`) — create, list, enable, disable, trigger, delete, view logs, and get stats programmatically.

See the [Tasks guide](/docs/mira/tasks.md) for no-code setup or the [Tasks API](/docs/sdk/core/tasks.md) for programmatic access.

**February 2026**

## User memory

Agents now remember user preferences, facts, and context across conversations:

* **Automatic memory** — agents learn and store important details (names, preferences, decisions) from natural conversation
* **Per-user isolation** — each user has their own private memory store, separate from agent training data
* **Memory bootstrap** — relevant memories are automatically injected into context at the start of each conversation
* **Vector-powered recall** — memories are semantically searchable, so the most relevant context surfaces automatically
* **Memory management** — view and delete per-user memories from the **Subscribers** page in the dashboard

## Context compaction

Long conversations no longer hit context window limits:

* **Automatic compaction** — when conversation context reaches 70% of the model's limit, older messages are summarized
* **Memory-safe** — important facts are extracted and saved to user memory before compaction
* **Model-aware** — compaction threshold adapts to each model's actual context window size
* **Transparent** — recent messages are preserved intact, only older history is summarized

**January 2026**

## Proactive trigger system

> [!NOTE]
>
> Removed in July 2026. Event Triggers and the trigger script have been retired.

Transform your agents from passive chatbots into proactive assistants that reach out at the right moment:

* **Smart triggers**: Fire events based on user behavior (errors, cart abandonment, page visits)
* **Multiple UI types**: Button, toast notification, banner, or inline embed
* **Customizable appearance**: Colors, text, positioning, animations
* **Delayed triggers**: Show help after a configurable delay

## Screenshot capture

Capture visual context when triggers fire to give your agent full situational awareness:

* **Automatic screenshots**: Capture the page when users click the trigger
* **Quality control**: Configurable JPEG quality and max dimensions
* **Privacy-first**: Opt-in only, users must enable per-trigger
* **Requires html2canvas**: Lightweight dependency (\~40KB gzipped)

## Multi-bot routing

Route different events to specialized agents on the same page:

* **Per-trigger botId**: Override the default agent for specific triggers
* **Specialized agents**: Sales bot for pricing, support bot for errors, billing bot for invoices
* **No re-initialization**: Switch between bots without calling init() again

## Rich context passing

Provide your agents with full situational context:

* **Source URL**: Agents know which page the user is on
* **Custom context**: Pass any data (cart value, error codes, user info)
* **User identification**: Associate triggers with logged-in users
* **Prompt templates**: Use `{{context.field}}` in agent instructions

**December 2025**

## Reasoning models support

* **OpenAI O-Series**: O1, O1 Mini, O3 Mini with extended thinking
* **Claude 3 Opus**: Extended thinking capabilities
* **Google Gemini**: 2.0 Flash Thinking mode
* **xAI Grok**: Grok 3 Reasoning
* **DeepSeek R1**: Cost-effective reasoning model
* Multiple reasoning modes: `auto`, `standard`, `agent`

## Expanded model catalog

* **50+ models** across 9 AI providers
* **GPT-5 series**: GPT-5, GPT-5.1, GPT-5-mini, GPT-5-nano
* **GPT-4.1 series**: GPT-4.1, GPT-4.1-mini
* **Claude 4.5 series**: claude-sonnet-4-5, claude-haiku-4-5
* **Gemini**: Gemini 3, Gemini 2.5 Pro, Gemini 2.5 Flash Lite
* **Grok**: Grok 4.1, Grok 4, Grok 3 variants

## New provider support

* xAI integration
* DeepSeek integration
* Existing: OpenAI, Google, Anthropic, Mistral, Cohere, Groq, Ollama

Learn more about [reasoning models](/docs/providers/reasoning-models.md) and [available models](/docs/providers/overview.md).

**December 2025**

## No-code agent creation

* **Visual dashboard interface**: Create agents without writing any code
* **Quick setup**: Get started in under 5 minutes
* **Custom branding**: Fully branded agent pages with custom domains
* **Multiple deployment options**: Embed widgets, standalone pages, or custom URLs

## Training made simple

* **File uploads**: Drag-and-drop PDF, Word, and text files
* **URL-based training**: Add website content with one click
* **Direct text input**: Paste content directly into the builder
* **Memory management tab**: Simplified interface for managing training data

## Lead capture & monetization

* **Email gating**: Built-in lead capture forms
* **94%+ conversion rates**: Optimized for maximum subscriber growth
* **Lead analytics**: Track and export subscriber data
* **Integration ready**: Connect with your existing CRM and marketing tools

## Custom branding & design

* **White-label experience**: Remove BoostGPT branding
* **Custom themes**: Choose colors, logos, and styling
* **Conversation starters**: Pre-configured example questions
* **Custom greetings**: Personalized welcome messages

Get started with the [Creator quickstart guide](/docs/quickstart.md).

**December 2025**

## MCP server generation

* **Auto-generate from OpenAPI specs**: Convert any OpenAPI specification to MCP server
* **Postman collection import**: Transform Postman collections into MCP servers
* **Managed hosting**: Custom MCP URLs at `https://mcp.your-server-id.boostgpt.co`
* **External compatibility**: Use with Claude Desktop, ChatGPT, and other MCP-compatible agents

## Pre-built connector library

**45+ ready-to-use connectors** including:

* **Calendar**: Calendly, Google Calendar, Cal.com, Outlook
* **Email**: Gmail, Outlook, SendGrid, Mailchimp, Twilio
* **CRM**: HubSpot, Salesforce, Pipedrive, Zoho
* **Productivity**: Notion, Asana, Jira, Linear, Trello, ClickUp, Monday.com
* **Storage**: Google Drive, Dropbox, OneDrive
* **Development**: GitHub, Vercel, Netlify
* **Analytics**: Google Analytics, Mixpanel, Datadog, Sentry
* **AI & Media**: Stability AI, Runway, Kling AI, Luma Labs, Unsplash, Pexels
* **E-Commerce**: Shopify, WooCommerce, Stripe
* **Communication**: Discord, Slack

Learn more about [MCP integrations](/docs/integrations/mcp-overview.md).

**December 2025**

## Multi-platform support

* **Discord**: Full Discord bot integration
* **Telegram**: Telegram bot adapter
* **Slack**: Slack workspace integration
* **WhatsApp**: WhatsApp Business API support
* **Crisp**: Live chat integration
* **Custom adapters**: Build your own platform adapters

## Custom adapter framework

* Standardized adapter interface
* Example Twitter adapter provided
* Event-driven architecture
* Message routing and context management

Explore the [Router SDK documentation](/docs/sdk/router/overview.md).

**December 2025**

## Event subscriptions

* `connector.created` - Connector creation events
* `connector.updated` - Connector configuration changes
* `connector.deleted` - Connector deletion events
* `subscriber.joined` - New subscriber capture
* `memory.created` - Memory source additions

## Security features

* HMAC SHA-256 signature verification
* `x-boostgpt-signature` header validation
* Secure event delivery
* Webhook testing with ngrok support

Learn more about [webhooks](/docs/developers/webhooks.md).

**December 2025**

## Advanced chat parameters

* **reasoning\_mode**: Choose from `auto`, `standard`, `agent`
* **temperature**: Control model creativity (0-1 range)
* **model override**: Change models per request
* **provider\_host**: Self-hosted Ollama support
* **provider\_key**: Bring Your Own Key (BYOK) for any provider
* **source\_ids**: Target specific training sources
* **memory control**: Toggle agent memory per request

## Response customization

* Real-time streaming with `stream` parameter
* Token limit control with `max_reply_tokens`
* Instruction override per request
* Conversation context with `chat_id`

Refer to the [Chat API documentation](/docs/sdk/core/chat-api.md).

**December 2025**

## Training capabilities

* **Multiple source types**: Text, websites, files, multiple URLs
* **Status tracking**: Monitor processing status (`processing`, `success`, `failed`)
* **Token usage**: Track tokens per training item
* **Source management**: Update and delete training data
* **Website crawling**: Train on entire domains

## Supported formats

* PDF documents
* Word documents (DOCX)
* Text files
* Markdown files
* CSV files
* PowerPoint presentations

Learn more about [training data management](/docs/sdk/core/training-data.md).

**June 2023**

## File upload functionality

* Added file upload capability to chatbot builder source management
* Users can now upload files directly within the source management feature
* Train your AI bot with local documents more efficiently

## Chat UI improvements

* Updated chat interface to display messages in bubbles
* Redesigned for more engaging and intuitive user experience
* Enhanced aesthetics and readability of chat feature

## Message retry functionality

* Implemented "failed to send. Retry" feature for chat messages
* Informative prompts when network issues prevent message delivery
* Option to retry sending failed messages
* Improved message delivery status visibility

## Network connectivity handling

* Event stream automatically stops when network connection is lost
* Prevents unnecessary data requests and reduces potential errors
* Smoother user experience during network interruptions
* Prevents data inconsistencies

**June 2023**

## Build your own AI Assistant

* Create fully personalized AI Assistants
* Customize and configure according to specific requirements
* Enhanced flexibility and versatility

## Zapier integration

* Automate chatbot workflows with Zapier integration
* Seamless data exchange between chatbot and other applications
* Streamlined processes and improved productivity

## Subscriber list growth

* Leverage AI Assistant to grow subscriber lists
* Facilitate seamless interaction and engagement
* Enhanced subscription management capabilities

Read more about these updates in our [newsletter](https://boostgpt.co/newsletter).

**June 2023**

## bot/subscriber/readall endpoint

* New API endpoint to fetch all subscribers associated with a bot
* Retrieve comprehensive list of subscribers in a single request
* Streamlined method for accessing subscriber data
* Simplified subscriber information retrieval for analysis, management, and reporting

Refer to the [API reference](/docs/api-reference/introduction.md).

**June 2023**

## bot/chat/readall endpoint

* Introducing chats pulling functionality for chatbots
* Effortlessly retrieve all chats in a bot
* Access valuable insights and information
* Enhance analysis and improve customer support
* Drive data-driven decisions

Refer to the [API reference](/docs/api-reference/introduction.md).

**May 2023**

## bot/chat/read endpoint

* Chat history pulling functionality for specific static chat\_id
* Effortlessly retrieve chat history
* Access past conversations for enhanced analysis
* Improve customer support with conversation insights
* Drive data-driven decisions

Refer to the [API reference](/docs/api-reference/introduction.md).

**May 2023**

## bot/chat/delete endpoint

* Enable chat history deletion for specific static chat\_id
* Users can reset conversations with the bot
* Enhanced user control over chat history
* Promotes fresh starts with the bot

Refer to the [API reference](/docs/api-reference/introduction.md).

**May 2023**

## Response streaming control

* Added "stream" parameter to bot/chat endpoint
* Customize response delivery behavior
* Choose between partial progress updates or complete response
* Set stream to `true` for real-time updates or `false` to wait for complete response

Refer to the updated [API reference](/docs/api-reference/introduction.md).

**May 2023**

## Conversational chat support

* Introduced chat\_id parameter to bot/chat endpoint
* Engage with AI bot in conversational manner
* Maintain context across multiple messages
* Enhanced user experience with continuous conversations

Refer to the updated [API reference](/docs/api-reference/introduction.md).

**May 2023**

## Multiple file link support

* Added support for multiple file links in bot training
* Updated source/create and source/update endpoints
* Train bots using wider variety of file sources
* Improved accuracy and effectiveness with diverse training data

### Supported file types

* CSV files
* PowerPoint presentations (PPT)
* Word documents (DOC)
* PDF documents
* Markdown files (MD)
* Text files (TXT)

Example request:

```json theme={null}
{
  "project_id": "13c8cc4a-ddef-11ed-b5a3-33d8a09a24e3",
  "bot_id": "bc65f4aa-e41a-11ed-ad91-a2ce1f33a089",
  "tags": ["terms", "privacy"],
  "type": "file",
  "source": [
    "https://www.example.com/path-to-your-file.csv",
    "https://www.example.com/path-to-your-file.ppt",
    "https://www.example.com/path-to-your-file.doc",
    "https://www.example.com/path-to-your-file.pdf",
    "https://www.example.com/path-to-your-file.md",
    "https://www.example.com/path-to-your-file.txt"
  ]
}
```

Refer to the [API reference](/docs/api-reference/introduction.md).

**May 2023**

## Multiple webpage link support

* Added support for multiple webpage links when training bots
* Updated source/create and source/update endpoints
* Enhanced training capabilities with multiple web sources
* Improved bot accuracy and response quality

Example request:

```json theme={null}
{
  "project_id": "13c8cc4a-ddef-11ed-b5a3-33d8a09a24e3",
  "bot_id": "bc65f4aa-e41a-11ed-ad91-a2ce1f33a089",
  "tags": ["property", "rent"],
  "type": "webpage",
  "source": [
    "https://www.propertyfinder.ae/en/rent/properties-for-rent.html?page=7",
    "https://www.propertyfinder.ae/en/rent/properties-for-rent.html?page=6",
    "https://www.propertyfinder.ae/en/rent/properties-for-rent.html?page=4",
    "https://www.propertyfinder.ae/en/rent/properties-for-rent.html?page=2"
  ]
}
```

Refer to the [API reference](/docs/api-reference/introduction.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.