Changelog

Improved attachment handling

• Refactored agent triggers to declare what attachments to expect
• Consistent service exception handling
• Update output governance policy handling
• Updated MCP tools and server descriptions
• Fixed counting for orgs with removed members
• Fix handling of errored evaluations

Bug fixes

• Fixed API reference page
• Fixed missing API docs
• Fixed plan change charge edge case

Bug fixes

• Fix alerts doc
• Fix auto-rollback
• Fix renewal case
• Fixed scheduled pull edge case

Improvements and bug fixes

• Agent creators choose whether attachment mismatches fail an email run and sources show rejected attachments
• Screening conversions of source files are reported as unattributed usage
• Implemented streaming of agent traces when agents are running
• Cache agent data
• Embed images inside documents
• Fix cancellation edge case
• Fix org accounts to show credit info for paying acocunt
• Fix upgrade paid after period ended
• Fix usage ID handling
• Fix usage recording of screening conversions
• Fixed doc inconsistencies
• Fixed query param issue
• Fixed usage recording of incomplete OpenAI calls

Improvements

• Clear step errors from purged runs

Improvements and bug fixes

• Option to stream through reasoning
• Fix low credit alert
• Fixed recursive lambda

Model updates

• Added Gemini 4 Argon
• Deprecated GPT 4 models

Model catalog improvement, bug fixes

• Add deprecated filter to model catalog and model selectors
• Added email rate limits
• Fix credit locale formatting
• Fix error text passed to models
• Fixed retention handling of observed governance evals
• Hardened email domain ownership verification
• Parity of media handling for websites
• Prevent tool errors from flagging MCP client auth needed
• Update handling of org invites

Multi-modal indexing, attachments UI, and bug fixes

• Index media on web pages
• Index RSS media
• Make attachments viewable
• Expanded governance evals to supported media
• Fix display of governance evals
• Fixed content source settings
• Hardened governance policy tests

Index images in documents

• Fixed indexing of images in docs

Bug fixes

• Don't re-embed embedded media
• Fixed ElevenLabs integration
• Fixed scan of email attachments
• Fixed user must select plan at end of trial
• Remove closed account from orgs

Demo/onboarding scheduling, model updates and bug fixes

• Added demo/onboarding booking links
• Deprecated models
• Fix hydration errors
• Fixed cache settings for ZAI GLM 5.3 Flash

Introducing email content sources

• Implemented email content sources
• Implemented indexing of non-text media
• Updated email templates
• Added Claude Haiku 5.5, Gemini Nano Banana 2.1, and Apodex 1.1 Mini
• Fixed empty streaming error

Bug fixes

• Fixed data sharing and non-streaming prompt call fallback
• Fixed slow response handling of ZAI GLM 5.3
• Fixed temperature field for Grok 4.7
• Improved timeout handling
• Updated terms to be more clear about abuse

Content source features

• Implemented governance evals on content source imports
• Implemented indexing of content attachmenents

Bug fixes

• Corrected gpt 6 luna token caps
• Fix cancel of early renewal
• Fix cancelation of model experiments
• Fix governance screening handling
• Fix model routing
• Fix transfer of file attachment when memory bank is transfered
• Fixed auto top up case
• Fixed prompt call compaction
• Fixed the agent AI assistant to be able to use all the trigger types

Bug fixes

• Added support for effort options to models that support it
• Fix file references in prompt calls
• Fix knowledge base deletion
• Fix references to files
• Fixed cloud drive triggers
• Fixed Gemini thinking token billing
• Renewal charge fix
• Updated list source contents description

Model updates

• Model deprecations

Doc search feature and model updates

• Add simple search in docs
• Added OpenAI GPT 6.1 Sol
• Added Inclusion AI Ling 3.1 Flash

Model updates and updated delinquent account handling

• Added Anthropic Claude Sonnet 5.5, and Xiaomi Mimo V2.6 Flash and Pro
• Close delinquent accounts after 60 days

Bug fixes

• Fixed decode turbo-stream response error
• Fixed failed subscription charge retry race condition

Step attachment improvements

• Improved handling of attachments

Bug fixes

• Added resize to prompt call field
• Fixed file attachments threaded through steps
• Fixed long context pricing
• Fixed storage blob access

New models and bug fixes

• Added canonical tags
• Added new models and improved price scraping
• Fixed agent trace for rejected email triggers
• Fixed conversation compaction
• Fixed custom alias for custom domains
• Fixed eval retry failure
• Fixed governance enforce cap
• Fixed GPT pricing
• Fixed org deletion
• Fixed binary attachments to send_email
• Updated multi modal outputs design

Bug fixes

• Email agents can't key conversation memory per email thread
• Clarify UI and documentation purpose of stripping quoted reply chains
• Fix temperature support for Opus 5 Sonnet

Bug fixes

• Fixed Claude 3.5 Haiku retirements
• Fixed delete org with transfered resources
• Fixed email agent trigger on duplicated agent
• Fixed simulate email
• Fixed unparseable redactor reply handling
• Implemented option to quote email threads
• Implemented UI to control quote removal feature
• Validated metadata references

Added feature to strip quoted replies in email chains

• Added feature to strip quoted reply chains in emails
• Excludes transferred connections from the former owner's existing-sources context
• Fix monthly overage budget truncation
• Removes redactor input cut

Bug fixes

• Eval failed when trace has an empty input
• Fixed email threading and match sender address
• Fixed org switching via MCP

SEO improvements

• Added robots.txt and sitemap.xml
• Enabled sitemap.xml in local dev #minor
• Fixed receiving emails on vanity sub-domains
• Ungrouped traces in Evals table, each it's own row my eval run

Bug fixes

• Added project review skill
• Fixed access and alerts for cancelled subscriptions until end of period
• Fixed brittle git tag
• Fixed discount code handling
• Fixed email domain reclaim notice
• Fixed subscription period charge that cannot be settled

Bug fixes

• Fixed cache discounts
• Fixed charge retry on plan upgrade
• Fixed govnernance blocking email from sending
• Fixed plan downgrade handling
• Fixed 1 hour cache billing
• Updated issue filing thresholds

Bug fixes

• Blocked alerts to offboarded org members
• Fixed delete with api key
• Fixed governance eval billing
• Fixed inbout email rejection alert email
• Fixed Stripe webhook signature verification

Bug fixes

• Fixed credit usage linking on evaluator outage
• Fixed ordering when reacting to payment status updates
• Fixed org API key check
• Persisted governance usage records

Bug fixes

• Delete prompt blobs after trace purge
• Do not quote source files with line numbers
• Updated to always check for PR when reviewing

New models and bug fixes

• Model updates
• Fixed org credit alerts
• Fixed upload to only org owners and admins
• Fixed alert unsubscribe
• Improved credit usage reporting
• Reconciled Stripe charges against our own records

Bug fixes

• Added Stripe idempotency key
• Fixed usage records for compaction step call
• Ma=de MCP client alerts mutable
• Reconciled Stripe records against our records
• Scoped embedding listing to account

Bug fixes

• Display billing and invoices for delinquent accounts
• Fixed credit formatting
• Fixed listing of embeddings with no embeddings
• Linked step usage IDs
• Resolved email template names at build time
• Updated website content source wizard

Bug fixes

• Prevented cross account MCP tools data transfer
• Show all credit types in the UI

Bug fixes

• Docs Claude Code MCP server missing /mcp at the end of url
• Fixed email greetings
• Fixed inferred usage use case
• Fixed trial conversion re-charge
• Fixed handling of bad SVGs
• Improved price and capability scraping
• Improved price scraping
• Prevent charactor from aborting send email
• Prevent double rendering of footers
• Prevent loss of usage IDs in tool loops
• Prevent MCP upload from writing to another account
• Prevent resubscribe from reporting payment failure

New models and bug fixes

• Added new models. Improved pricing collection
• Create new content versions when content is updated
• Fixed Anthropic output cap handling
• Fixed state of subscriptions for renewal
• Fixed recipient of subscription emails
• Fixed resubscribe edge case
• Linked ancillary tool usages to prompt call records

Display LLM input and output caching results in agent step traces

• Display caching in agent step traces
• Report retryable failures to Sentry

Bug fixes

• Fixed usage ID handling in the tool-loop test harness
• Fixed Claude cached token billing
• Fixed Pinecone retrieval bug
• Madde sure usage ID is added
• Prevented delete_vectors from requesting empty delete

Bug fixes

• Fixed billing of cached input tokens
• Fixed ranking and sorting results as similarity scores rather than di…

Bug fixes

• Fixed unbounded alt text from scraped pages
• Consolidated rate resolution
• Cleaned up usage rate keys
• Fixed content access account hardening

Withdrew the unimplemented step caching fields

• Removed `extended_caching_days` and `cache_all_minutes` from the Prompt Call and Retrieval steps. Neither field was ever honoured — no step result caching exists — so setting one changed nothing. Saved workflows still load; the dead keys are dropped on the next save.

Documentation: calling Seclai from a browser

• New "Calling from a Browser" documentation page explaining why the API sends no CORS headers, and how to build a browser-facing app: a backend proxy that exposes a small allowlist of named operations rather than forwarding paths.

Added new models

• Added Gemini 3.7 Flash and Ternary Bonsai 27B.
• GPT Image 2 is now available for image generation steps.

Added new models

• Added Grok 4.6, Inkling Small, Macaron V1 Tall, Muse Glimmer 30B, Qwen3.8 Max, and Qwen3.8-2.4T-A95B.
• Deprecated the media generation models retiring on 2026-08-19.

Onboarding improvements

• Added a "How did you hear about Seclai?" question to onboarding.

Unified AI assistant panel

• Added a unified AI assistant panel for streamlined workflow creation

Granular endpoint versioning

• Added support for granular endpoint versioning
• Made all paginated list responses consistent

Added new model

• Added Claude Opus 5

Added email vanity domains and new models

• Added support for email vanity subdomains and custom domains
• Added Gemini 3.6 Flash and Gemini 3.5 Flash Lite models

Added new models

• Added Kimi K3 and Thinking Machines Inkling

Added email agent trigger and new models

• Implemented email agent trigger
• Added GPT-5.6 and Grok 4.5 models

Dropbox integration

• Integrated Dropbox cloud drive

Agent editor improvements

• Updated the agent visual editor with zoom and pan support
• Move agent import from modal to dedicated page

Added new models

• Added Anthropic Claude Sonnet 5 and ElevenLabs models

Added new models

• Added Z.ai GLM 3.2 and other open-weight models.

Agents can generate images, audio, and video

• Added step types and tools for image/audio/video-only LLMs
• Implemented rate limiting of webhook calls per host and account

Slack bot and prompt call improvements

• Added a Slack bot MCP server for connecting agents to Slack
• Added capability filters to the model picker in Prompt Call steps
• Polished the UI for adding new MCP clients
• Fixed a bug in the evaluation step

MCP and governance fixes

• Fixed the MCP OAuth callback flow
• Governance re-runs on agent traces now respect whether governance is enabled

Wait steps and new models

• Added a wait step type for time-based delays in agent workflows
• Added Anthropic Fable 5
• Fixed the credit rate display

MCP tool calls for agents and models

• Added a catalog of ~250 preconfigured MCP servers for one-click client setup
• Added an MCP tool call step type for agent workflows
• Enabled any tool-calling LLM to invoke MCP tools during a run

JSON and XML schema validation

• Added support for JSON and XML schema validation to the extract content step

Added new models

• Added Anthropic Claude Opus 4.8
• Added MiniMax M3
• Added Xiaomi Mimo V2.5

Multi-modal agents and knowledge bases

• Agents and knowledge bases can now receive, process, and generate images and files in addition to text

UI improvements

• Add + icon to all create/add buttons across tabs
• Add soft-delete support for Model Experiments (API + UI actions menu)

UI improvements

• Align Governance tab selection with nested route patterns
• Align HITL review “waiting on me” toggle to left and keep filters right-grouped

UI consistency and empty state improvements

• Added pending (HITL) traces to Dashboard summary cards and chart
• Show OAuth info banner in API Keys empty state
• Refined empty states, dashboard cards, and tab styling across the app
• Moved Import/Export to overflow dropdown on Agent details page

Model updates and feature enhancements

• Implemented human in the loop (HITL) execution
• Added model updates
• Added relative time string substitution options
• Fixed handling of recurring payments

Added human in the loop feature

• Added human in the loop step
• Added page for managing human in the loop tasks

Added if/else and switch steps

• Added if/else and switch steps

Agent step updates and AI assistant improvements

• Added for each step type
• Split memory bank steps into conversation and general memory bank steps
• Improved solution and agent AI assistants

Model updates and settings UI improvements

• Add Gemini 3.5 Flash, Grok Build 0.1, Ring 2.6 1t; deprecate 5 xAI vendor models
• Update settings UI

UI improvements

• Update Memory Banks UI
• Update Organizations UI

Side navigation update

• Update side nav again

Updated the model experiments tab

• Make updates to Model Experiments tab

Governance policy review feature

• Governance policy review feature

Governance evaluation retention and model updates

• Governance evaluation retention
• Added Gemini 3.1 Flash Lite, Xiaomi Mimo V2.5 Pro

Governance latency optimizations

• Governance latency optimizations

Redaction governance policy mode

• Redact governance policy mode
• Simplify Agent page

Import agents

• Implemented the ability to import exported agents

xAI Grok 4.3

• Added xAI Grok 4.3 model

DeepSeek V4 Flash and Pro

• Added DeepSeek V4 models

OpenAI GPT-5.5 and GPT-5.5-Pro

• Added GPT-5.5 and GPT-5.5-Pro

Blog and web tools

• Added blog
• Implemented web fetch and web search steps and tools

Qwen 3.6 35B A3B

• Added Qwen 3.6 35B A3B

Moonshot Kimi K2.6

• Added Moonshot Kimi K2.6

Promo code support

• Implemented basic promo code support

Agent trace pseudo-steps

• Show prompt scans and evals as steps in agent traces

Unified content source types

• Combined the file upload and custom index content source types

GitHub login

• Added log in with GitHub

Knowledge base chat and model experiments

• Implemented knowledge base chat
• Implemented streaming of model experiments

Streaming output

• Implemented streaming output

Waitlist removal and logout

• Disable waitlist
• Implemented log out

Prompt Scanner

• Launched ML-based prompt scanner that blocks prompt injection attacks at all platform ingress points
• Updated terms and conditions to note input scanning

Agent optimizations and governance UX

• Optimized agent runs
• Added usage tabs to governance and governance policy detail pages
• Only show governance options for steps that support governance
• Improved the agent AI assistant
• Updated agent delete confirmation modal to have user enter the agent name

Resource exports and credit usage tracking

• Added time frame selection to knowledge base and content source content
• Export resources
• Record credit usage for agent eval, governance and model experiments

Governance evaluation for agent traces

• Added ability to test with agent traces for existing governance policies
• Added missing ability to run governance evals on all agent traces and…

OAuth API documentation

• Update API doc to include Oauth

Account API and endpoint updates

• Added endpoint to obtain account IDs
• Updated API spec and docs
• Updated me endpoint return type

OAuth support and streaming updates

• Added streaming updates when making accepted changes
• Dropped proration of monthly credits on upgrade
• Implemented OAuth for consumer APIs and MCP

In-process data caching

• Implemented caching of frequently used data

Agent performance and governance launch

• Optimized agent latencies
• Enabled the governance feature

Credit alerts and email notifications

• Real-time credit alert evaluation after every deduction
• Email notifications for model deprecation, sunset, and new releases
• Email preference management in account settings
• One-click email unsubscribe support

Content source embedding migration

• Implemented content source embedding migration

Model auto-upgrade and observability

• Implemented model auto-upgrade feature
• Implemented model-centric observability

Model playground from catalog pages

• Added an account-level models page with LLM playground testing for simple prompts and JSON payloads
• Added public model catalog and model detail buttons that route into the authenticated account playground

Input styling consistency

• Standardized text-like inputs and field wrappers for consistent behavior with dropdowns

Getting started docs refresh

• Added pictures to the docs getting started guide

Agent evaluation improvements

• Implemented agent evaluation improvements

Trace navigation improvements

• Added a back button in traces to return to the previous nested screen

Model recommendations and prompt updates

• Implemented recommendations for new, deprecated, and sunset models
• Updated the AI assistant system prompt

Content export and model updates

• Implemented content source export
• Added Agent ID to the details page
• Added new models

AI assistant examples and pricing updates

• Added AI assistant examples
• Adjusted pricing, credits, and rate limits

Account access fixes and governance follow-ups

• Fixed top-level pages for delinquent accounts
• Governance follow-ups
• Updated frontend packages

AI assistant UX and navigation fixes

• Disabled return key submission in AI assistant modals
• Fixed solution and agent AI assistants

Governance workflow and compaction updates

• Added a scheduled task to run memory bank compaction
• Added context compaction step and prompt call option
• Handled RSS feeds with private links
• Implemented sample governance policy previews and added AI assistant support

Changelog tooling and platform fixes

• Implemented changelog documentation and updated tools
• Made recipient user ID optional
• Prevented markitdown from loading in lambdas

Memory banks & model updates

• Implemented memory banks
• Deprecated Gemini 3 Pro
• Implemented downgrade resource management

Gemini 3.1 Pro support

• Added Gemini 3.1 Pro

File-triggered dynamic input agents

• Implemented file upload to trigger dynamic input agents

New AI models & UI refinements

• Added OpenAI GPT 5.3-Codex
• Fixed creating and updating agents via solution AI assistant
• Make Step colors category colors
• Make text area box in AI Assistant adjustable
• Move Organizations from left nav bar to settings tab
• Fixed table scroll container and layout rendering

Content steps & AI assistant feedback

• Added add content step
• Change order of Sources and Agent in Solutions pipeline visual
• Fix file uploads in production
• Implemented AI assistant feedback feature
• Solution AI fixes

Solutions AI assistant & attachments

• Added write metadata step
• Completed the solutions AI assistant
• Added write metadata, write content attachment, and load content attachment steps
• Fixed bugs and refined solutions AI assistant and attachment handling
• Made sure attachments are private
• Normalize whitespace on metadata/attachment keys before validation
• UX tweaks

Retry step support

• Implemented retry step

Governance system & solution AI assistant

• Added insight step
• Added call agent step
• Added ability to toggle speed vs thoroughness of evaluations
• Added UI for managing governance settings
• Implemented AI solution assistant and solutions
• Implemented governance system
• Implemented handling of large input in agent steps
• Linked governance evaluations to agent traces
• Renamed run history to traces
• Added feature-flagged governance POC
• Updated file-storage version

Observability and performance monitoring

• Added agent run history duration graph

Agent AI assistant & parallel testing

• Added AI assistant for creating and updating agents
• Implemented run-to-step for step-by-step agent debugging

AI assistants for all steps & documentation

• Added AI assistants to all step types
• Documented alerts, API keys, dashboard, and resource transfers
• Updated landing page
• Implemented test agent run steps with historical run input
• Fixed sorting appearance in lists
• Fixed content source website wizard error when selecting 1 week

New AI models & documentation improvements

• Added Claude Sonnet 4.6 and new open-weight models
• New documentation pages for knowledge bases, content sources, organizations, and embedding models
• Improved getting started guide and credits & usage documentation

Agent templates & UI improvements

• Ability to create agents from templates
• New documentation site with introduction and getting started guide
• Improved agent definition page layout and navigation