Skip to main content

Unstract Cloud Changelog - 2026

v0.181.0​

Release Date: September 14, 2026

New Features​

  • Header mapping in the Agentic Table Extractor: A new Enable Header Mapping switch in the Agentic Table Extractor settings makes an extraction also report, for every field you asked for, the source column in the document it was read from — useful for reviewing an extraction or tracing a value back to the original table. It's off by default. Each reported column is checked against the columns the document actually contains, so a field is left blank rather than pointed at a column that isn't there.
  • Text extraction time in execution metrics: When you request metrics from an API deployment execution, the response now reports how long text extraction took, alongside the indexing time it already reported. Extraction is often the largest part of a run on big documents, so a slow execution is far easier to account for.
  • Look up the organization behind a platform API key: The Platform API now answers "which organization is this key for?" at /api/v1/unstract/whoami/, returning the organization's ID and name along with the platform API key's name and permission tier. Scripts and command-line tools no longer have to copy the organization ID out of a browser URL before they can reach the organization-scoped endpoints. The endpoint that lists an organization's API deployments is now published in the OpenAPI specification too, so generated clients and tooling pick up both.
  • Dashboard totals cost less to keep current: Hourly figures on the dashboard refresh exactly as often as before. Daily and monthly totals now recompute once an hour instead of every 15 minutes, so they can sit up to an hour behind the newest activity — in exchange for a large reduction in the background work behind the dashboard.

Bug Fixes​

  • Long multi-page tables extract more accurately: A set of Agentic Table Extractor fixes for documents whose tables run across many pages. A row that continues over a page break now merges back into a single row instead of leaving a stray partial row at the top of the next page; summary and totals pages that fall outside a table's own page range are no longer skipped, so their values come through; very dense pages that previously came back empty now retry and return their rows; and rows from nested sub-tables keep the link to the parent row they belong to. A malformed nested entry is now skipped instead of ending the whole extraction.
  • Webhook addresses are validated before they're called: A webhook URL is now checked when you save a notification and again on every delivery, with the same check applied on both the notification and post-processing paths — so an address that can't safely be called is refused as you enter it, rather than failing quietly at delivery time. Deliveries also no longer follow redirects to a different destination.
  • Prompt Studio records stay inside your organization: Requests that name a Prompt Studio document, prompt, profile or output directly by its identifier are now confirmed to belong to your organization before they are acted on.
  • Markup is rejected in names and descriptions: Names, descriptions and similar short text fields across the product now refuse HTML markup and script-like content at the point you submit them, instead of storing it as typed. Fields that legitimately carry such content — your prompt text, and model responses — accept it exactly as before.

v0.180.0​

Release Date: September 8, 2026

Bug Fixes​

  • Every organization is reachable on the organization picker: If you belong to more than about five organizations, the list on the organization selection screen was cut off at the edge of the screen with no way to scroll, leaving the remaining organizations unreachable, and the Unstract logo above it collapsed out of view. The list now scrolls and the whole screen stays intact however many organizations you belong to.

v0.179.0​

Release Date: September 2, 2026

New Features​

  • Refreshed interface styling: Buttons, labels and icons across Unstract now render with the platform's new visual design, so screens look consistent whichever part of the product you're in. Layouts, navigation and what each control does are unchanged — nothing has moved and nothing you use today works differently.

Bug Fixes​

  • Image challenges can be completed inside dialogs: When a form opened in a dialog asked you to solve an image challenge to prove you're not a bot, the challenge tiles couldn't be clicked, leaving no way to finish the verification. The challenge now responds to clicks as expected.

v0.177.0​

Release Date: August 20, 2026

New Features​

  • Newer Amazon Bedrock models: You can now use the OpenAI GPT-5.x, Google Gemma and xAI Grok model families hosted on Amazon Bedrock. Enter the model ID exactly as you do today — the adapter now sends these families to the endpoint that serves them, instead of failing. Guardrails and application inference profiles aren't available for these particular models and are ignored if you've configured them.

Bug Fixes​

  • Shared users can see execution logs: Teammates who receive an API deployment or pipeline through a group share, or through an organization-wide share, now see its runs on the Logs page — the page previously came up empty for them. At the same time, a run's logs and per-file results can only be opened by people the deployment or pipeline was actually shared with.
  • Model names appear on shared Prompt Studio projects: Prompt Studio output tiles now show the model name and icon for everyone who can open the project, not only its owner. Shared users previously saw a blank label and no icon.
  • Connector names in Agentic Prompt Studio settings: Project settings in Agentic Prompt Studio now show the names of the connectors in use instead of leaving them blank.
  • LLMWhisperer text extraction honours the line splitter strategy: The line splitter strategy selected on the LLMWhisperer Text Extractor-v2 adapter was never actually applied — every extraction silently ran with the default. The setting now takes effect, and extractions started from Unstract show the document's file name in LLMWhisperer usage reports instead of a generic placeholder. If you had saved a value outside the supported choices, extraction now reports an error rather than quietly ignoring it.
  • Extended Thinking works beyond Claude: The "Enable Extended Thinking" option now applies to non-Claude model families on Amazon Bedrock, where it previously had no effect at all.
  • Clearer failures on single-pass extraction: When a model answers a single-pass extraction with prose or a list instead of structured output, the run now reports a clear, actionable error instead of failing with an unexpected server error.
  • API keys are no longer written to logs: The LLMWhisperer and API Hub portals no longer record customer API keys in plain text in their request and key-management logs.

v0.176.0​

Release Date: August 12, 2026

New Features​

  • Global API deployment keys: You can now create a single API key that authenticates against several API deployments at once — either a chosen set of deployments or every deployment in your organization — instead of managing a separate key per deployment. Global keys live under Settings > Global API Deployment Keys, where you can create, edit, rotate, and revoke them.
  • Hosted MCP server: Unstract now hosts an MCP endpoint alongside every API deployment, so coding agents and other MCP clients can call Unstract as tools instead of building HTTP requests by hand. It sits at the deployment's existing URL with /mcp appended and authenticates with that deployment's own API key. Nothing extra to install, and no new credential to mint.
  • Unpublish a Prompt Studio tool: You can now remove an exported tool from the tool registry while keeping its Prompt Studio project — previously the only way to remove one was to delete the whole project. Tools still in use by a workflow or deployment are protected, and the refusal names what is holding on to them. Note that re-exporting later creates a new tool rather than restoring the old one, so anything referencing the previous tool needs updating.
  • Sortable, searchable list pages: The Prompt Studio, adapter, connector, and workflow list pages now share one layout with an Owned By column, sorting by name, created, or modified date, and search that matches either the resource name or its owner's email. Pages load one at a time and the picker dropdowns that select these resources now load past the first page too, so everything stays reachable as your organization grows.
  • Setup documentation links everywhere: Every Add Connector and Add LLM, Vector DB, Embedding, or Text Extractor screen now shows a link to that item's setup page, so you no longer have to leave the product and search for it. Items without a dedicated page link to their category overview.
  • Agentic table extraction improvements: Agentic Prompt Studio table extraction continues to get more accurate on long, multi-page tables, with better handling of per-group summary rows.

Bug Fixes​

  • Deploying an exported Prompt Studio tool works with LLM Challenge on or off: Deployments used to fail validation at the very last step — after everything else had succeeded — because the challenge model resolved during export was discarded before validation. The resolved model is now carried through, so deploys pass whether LLM Challenge is enabled or disabled, and a model you set for a specific project is no longer overwritten by your account-wide default.
  • Prompts fall back to the project's default LLM profile: Running a prompt that has no profile of its own now uses the project's default LLM profile, matching how indexing and single-pass extraction already behaved, instead of failing. The related error message now names both places a profile can come from, so it points at something you can actually change.
  • Correct tool picked when creating an API deployment from Prompt Studio: Creating an API deployment from a Prompt Studio project now matches the exported tool to its source project directly, rather than by name — so organizations with two projects sharing a name no longer risk deploying the wrong one.
  • API keys can only be created by the deployment's owner: Creating a key for an API deployment or pipeline is now checked against ownership on every path, closing a gap where another member of the same organization could mint a working key for a deployment they did not own.
  • Simpler API key creation requests: Creating a key through the API no longer requires repeating the deployment or pipeline identifier in the request body when it is already part of the URL, and a request naming a different target than the URL is now rejected instead of quietly picking one.
  • Multi-page and grouped table extraction fixes: Rows that continue across a page break now merge correctly when the continuation is the first row on the page, tables with only a single group now return their summary line, rows that lose their group association are re-attached instead of dropped, and vision-mode extractions retry automatically when output is cut short.

v0.175.0​

Release Date: July 29, 2026

New Features​

  • Grouped table output in Agentic Prompt Studio: Agentic table extractor table extraction can now group extracted rows by a section value — for example, nesting line items under each property or invoice — and optionally return a summary line of totals for each group. Grouping is now driven by the shape of your project's output schema instead of the separate "Group Key(s)" setting, which has been removed, and it runs as part of the main extraction pass rather than an extra pass over every page, so grouped extractions finish faster and cost less.
  • Lookups now work with agentic table prompts: You can now assign a lookup to an agentic table prompt to enrich its extracted table — previously lookups couldn't be attached to these prompts at all. By default the lookup runs once over the whole extracted table; a new "Enrich each row separately" option runs it per row for finer-grained enrichment, with a cost warning up front since it means one call per row. Rows that fail to enrich keep their original values and are reported in a summary instead of failing the run.
  • Faster Workflows page for large organizations: The Workflows list now loads a page at a time and searches by name across all your workflows, so the page stays quick as your workflow count grows.

Bug Fixes​

  • API Hub usage is recorded again: API Hub usage had stopped being recorded, so the subscription dashboard no longer reflected the pages your API calls processed, with no error shown anywhere. Usage is now tracked reliably for every completed call, including on shared API Hub products where the subscriber differs from the account that owns the underlying deployment.
  • Reasoning models work in agentic table extraction: Agentic table extraction no longer fails when the project uses a reasoning model that rejects the temperature setting.
  • Fewer split rows in multi-page tables: A row that continues across a page break is no longer split into separate records when the model returns the same identifier with different capitalization, so multi-page tables extract as single, complete rows.
  • List page columns line up: The "Owned By" and "Updated" columns now sit at the same position on every row across the Prompt Studio, adapter, connector, and workflow list pages, and list titles are no longer thinner than those in the Agentic Prompt Studio project list.

v0.174.0​

Release Date: July 23, 2026

New Features​

  • Group-key extraction for agentic tables: The agentic table extractor can now organize extracted table data under group keys, including for unlabeled, section-level fields, and supports multiple group keys within a single table. A dedicated extraction agent handles grouped fields and multi-sheet sample selection, improving accuracy when pulling structured data from documents and spreadsheets that repeat the same fields across several sections.
  • Co-owners for shared projects: You can now assign co-owners to Agentic Prompt Studio projects and other shared resources, so more than one person can manage and edit them.
  • MiniMax LLM support: You can now connect MiniMax models as an LLM adapter.
  • Created and modified dates on list pages: Adapter and Prompt Studio list pages now show when each item was created and last modified, making it easier to spot your most recent work.

Bug Fixes​

  • Correct ownership checks in Prompt Studio: Ownership permissions are now validated against the person making the request, so access to shared Prompt Studio projects is granted correctly.

v0.173.1​

Release Date: July 15, 2026

New Features​

  • Mistral embeddings: You can now use Mistral's mistral-embed model as an embedding adapter, giving you another option for indexing and semantic search.

Bug Fixes​

  • Newer Claude models work reliably: Claude Sonnet 5 and newer Claude models no longer fail on the deprecated temperature setting, so prompts using these models run without errors.
  • Output Analyzer shows profile names: The Output Analyzer tabs in Prompt Studio now display each profile's name instead of the underlying LLM name, making it easier to tell your profiles apart.
  • LLMWhisperer dashboard number formatting: Large numbers on the LLMWhisperer dashboard now include comma separators, so they're easier to read at a glance.
  • LLMWhisperer playground layout: Sample card descriptions no longer overlap the "Mode used" label in the LLMWhisperer playground.

v0.173.0​

Release Date: July 13, 2026

New Features​

  • Buy LLMWhisperer through AWS Marketplace: LLMWhisperer buyers who arrive from AWS Marketplace are now onboarded directly into the LLMWhisperer product and land on a dedicated marketplace page. Since billing is handled through AWS, these accounts see an AWS Marketplace billing notice in place of the standard Stripe pricing.

Bug Fixes​

  • More reliable tool execution: Fixed an issue in the tool runner that could cause tool runs to fail unexpectedly.

v0.172.0​

Release Date: July 6, 2026

New Features​

  • Restrict adapter and connector creation to org admins: A new organization setting lets admins restrict who can create LLM adapters and connectors to organization admins only. When it's on, non-admin members work from the shared adapters and connectors an admin has already set up. The toggle lives in Platform Settings and is off by default, so existing behavior is unchanged until you enable it.
  • Rotate platform API keys via the API: Platform API keys on the full_access tier can now call the key rotation endpoint, so credentials can be rotated through automation instead of only from the admin UI.
  • See your running platform version: Your profile page now shows the platform version you're currently on, making it easy to reference the exact version when reporting an issue.

Bug Fixes​

  • API deployment runs no longer get stuck on staging failures: When an API deployment run fails while staging its input file, the execution is now marked as failed with the error reason instead of appearing to run indefinitely.

v0.171.2​

Release Date: July 4, 2026

New Features​

  • AWS Marketplace onboarding for LLMWhisperer: The AWS Marketplace claim flow now covers the LLMWhisperer listing, with a dedicated claim landing page. Subscriptions are provisioned automatically the moment you claim them, so marketplace onboarding is fully self-serve.
  • Platform version on your profile: The profile page now shows the current Unstract platform version, so you can easily see which version you're running.

Bug Fixes​

  • LLMWhisperer usage on the dashboard for custom plans: Custom plan subscribers now see their LLMWhisperer usage on the dashboard, matching the visibility already available on other plans.

v0.171.1​

Release Date: July 1, 2026

Bug Fixes​

v0.171.0​

Release Date: July 1, 2026

New Features​

  • Excel support for agentic table extraction: The agentic table extractor now works with Excel workbooks (.xlsx and .xls), including multi-sheet files where each sheet is extracted on its own.
  • Automatic welcome for marketplace buyers: New customers who purchase through the marketplace are now welcomed automatically as soon as their subscription activates, with more reliable account claiming during checkout.

Bug Fixes​

  • Lookups: Fixed an issue where an orphaned assignment could block deleting a lookup, and switching the extractor now correctly re-runs extraction.
  • Reliable API updates: Updating an existing record through the API no longer fails with a false "already exists" error, so upsert-style requests succeed as expected.
  • Faster page loads: Parts of the app now load on demand, reducing the initial load time when you navigate between sections.

v0.170.2​

Release Date: June 24, 2026

Bug Fixes​

  • Long S3 presigned URLs are now accepted: URL fields no longer reject very long links, so long Amazon S3 presigned URLs validate and save successfully instead of being flagged as invalid.

v0.170.1​

Release Date: June 24, 2026

Bug Fixes​

  • Single-pass extraction handles non-object results: Document processing with single-pass extraction no longer fails when extraction returns a value that isn't a structured object, so these runs complete successfully instead of crashing.

v0.170.0​

Release Date: June 24, 2026

New Features​

  • Prompt Studio project transfer keeps per-prompt settings: Transferring a Prompt Studio project now carries each prompt's table and agentic extraction settings along with it, so a transferred project behaves the same as the original without manual reconfiguration.
  • Claim AWS Marketplace subscriptions by email: Customers who subscribe through the AWS Marketplace can now claim their purchase using their verified email address, streamlining marketplace onboarding.

Bug Fixes​

  • Google Drive connector: Several Google Drive fixes — the folder browser no longer breaks when a folder contains Google Docs files, and Google Drive connections now work reliably across all execution workers.
  • Adapters in use are protected from deletion: You can no longer delete an adapter that is still used by a workflow tool instance, preventing workflows from breaking unexpectedly.
  • Correct LLMWhisperer adapter key label: The API key field on the LLMWhisperer adapter is now labeled "LLMWhisperer Key" instead of the misleading "Unstract Key".
  • Org admins and service accounts see all lookups: Organization admins and service accounts now have visibility into every lookup in the organization.

v0.169.0​

Release Date: June 17, 2026

New Features​

  • Subscribe through the AWS Marketplace: You can now purchase and manage your Unstract Cloud subscription through the AWS Marketplace. A new marketplace landing and claim flow connects your marketplace purchase to your account, and usage is metered and billed automatically.
  • Smarter execution notifications: Notifications can now be batched within a time window so you receive a single grouped alert instead of one per run, and a new "notify on failures only" option lets you stay quiet on successful executions and hear about problems.
  • Platform API support for org-to-org cloning: The Platform API now supports service-account group management and name-based filters, making it easier to programmatically copy resources from one organization to another.

Bug Fixes​

  • Region switcher on the sign-up page: The region switcher now appears on the default sign-up page, so you can choose your data region when creating an account.
  • Social login for accounts with a shared email: Signing in with a social account that uses the same email as an existing account now resolves to the correct account instead of failing.
  • Safeguard for the last organization admin: The last remaining org admin can no longer demote themselves, so an organization can never be left without an administrator.
  • Sharing for exported Prompt Studio tools: Project shares now apply to exported Prompt Studio tools, and the share dialog correctly keeps its organization-wide share setting.

v0.168.0​

Release Date: June 11, 2026

New Features​

  • User groups and group-based sharing: You can now organize users into groups and share projects and resources with an entire group instead of inviting people one at a time. Group-based sharing works across Prompt Studio, Agentic Prompt Studio, and workflows, and adapters shared with a group can be used directly in Prompt Studio and workflows.
  • NVIDIA Build and OpenRouter adapters: Added OpenAI-compatible LLM and embedding adapters for NVIDIA Build and OpenRouter, expanding the set of model providers you can connect.
  • Automatic default LLM profile in Prompt Studio: New Prompt Studio projects now start with a default LLM profile created for you, and the document upload tooltip is clearer about what to do next.
  • Clearer adapter configuration forms: Adapter setup forms now display their fields in a defined order, with the Amazon Bedrock adapter surfacing its authentication fields more prominently.
  • Quick access to execution logs: Running a workflow now shows an alert with a direct link to its execution logs, so you can jump straight to the run details.

Bug Fixes​

  • Cleaner sharing dialogs: Organization admins no longer show up as noisy, redundant entries in resource sharing lists or the share dropdown.

v0.167.0​

Release Date: June 1, 2026

New Features​

  • AWS Bedrock Guardrails: The Amazon Bedrock LLM adapter can now apply an AWS Bedrock Guardrail to its requests, letting you enforce content filtering and safety policies on prompts and responses directly from your adapter configuration.

v0.166.0​

Release Date: May 28, 2026

New Features​

  • Full document text in API responses: API deployment responses can now return the full extracted text of the input file alongside the structured output, so you no longer need a separate call to retrieve the source text.
  • Request ID on error notifications: Error notifications now surface a request ID that ties the failure back to the originating request, making it faster to track down and report issues to support.

Bug Fixes​

  • Apostrophes in Platform API Key names: You can now use apostrophes in the name and description of a Platform API Key, so names like "Acme's key" are no longer rejected.

v0.165.0​

Release Date: May 27, 2026

New Features​

  • Agentic Prompt Studio onboarding: New users now land in a guided onboarding flow for Agentic Prompt Studio — a sample project, guided empty states across each step, and a dashboard CTA help you reach a working extraction faster.

Bug Fixes​

  • Large file uploads in Prompt Studio: File uploads in Prompt Studio now stream directly to remote storage instead of being buffered in backend memory, so larger documents no longer fail with memory pressure on the backend.
  • Org admin access to all organization resources: Org admins now have full access to every resource in their organization — Prompt Studio projects, adapters, and shared profiles — without each resource needing to be explicitly shared with them.

v0.164.0​

Release Date: May 22, 2026

New Features​

  • OpenAI-compatible LLM adapter: A dedicated OpenAI-compatible LLM adapter lets you connect any provider that speaks the OpenAI API (including self-hosted gateways and OpenAI's own gpt-5 / o-series models) without forcing it through the standard OpenAI adapter.
  • Prompt Studio HITL feedback indicator: Prompt Studio now surfaces a change indicator when a prompt's output has been edited via the HITL feedback loop, making it easier to spot prompts that need tuning.

Bug Fixes​

  • Table extraction — cross-page orphan rows: Subtable values placed by the LLM into the main values payload are now correctly merged into the parent row across page boundaries, fixing orphaned rows on multi-page tables.
  • Default adapters for invited users: Invited users now inherit the organization's default LLM, embedding, and vector DB adapters on first login, so they can start building workflows without manual adapter setup.
  • New-user signup hardening: Improved error handling during new-user signup.

v0.163.0​

Release Date: May 18, 2026

New Features​

  • Lookups V2 — post-extraction LLM enrichment for Prompt Studio: A new enrichment stage runs additional LLM lookups on already-extracted data, letting you cross-reference and normalize values without a second extraction pass. You can upload lookup reference files, share lookups with teammates (with sharing-notification emails), and review aggregated lookup metrics under a single lookup_llm list. See Look-Ups for details.
  • AWS Bedrock bearer token authentication: Bedrock LLM and embedding adapters now support long-term AWS_BEARER_TOKEN_BEDROCK authentication — a simpler alternative to static keys or IRSA for environments that prefer a single, long-lived API key. (Short-lived/refreshing tokens are not supported.)
  • Workflow execution log export: The workflow execution logs modal can now export logs as CSV or JSON for offline analysis and sharing.
  • full_access permission tier for Platform API: A new permission tier authorizes DELETE operations on the Platform API, letting admins explicitly grant destructive access without overprovisioning broader roles.

Bug Fixes​

  • Tool-run logs: Cloud plugin tool-run logs now stream into the workflow execution UI; output was trimmed and clarified, and logs now persist correctly even when no UI subscriber is active.
  • Adapters: Deprecated sampling parameters are stripped automatically for Claude Opus 4.7, and the Bedrock bearer token is no longer lost on adapter re-validation during extraction.

v0.162.0​

Release Date: May 11, 2026

Bug Fixes​

  • Worker observability: The next-generation worker executor is now wrapped with OpenTelemetry auto-instrumentation, so worker spans show up alongside backend traces for end-to-end visibility into workflow execution.

v0.161.0​

Release Date: May 7, 2026

New Features​

  • Agentic table extractor: A new multi-agent, LLM-powered table extraction capability for complex tabular documents in Prompt Studio, improving accuracy on structured tables that traditional extractors struggle with.

Bug Fixes​

  • S3 connector cross-region buckets: The S3 filesystem connector now drops cross-region buckets from the picker, avoiding listing errors during connector setup.
  • Multi-tenant subdomain CORS: SocketIO and Django now accept wildcard subdomain origins, making multi-tenant subdomain deployments work without per-tenant CORS configuration.

v0.160.1​

Release Date: April 27, 2026

Bug Fixes​

  • Human-in-the-Loop (HITL) audit UI: JSON fields now show edit markers with per-leaf diffs, and added vs. removed rows in list-valued fields are distinguished correctly. See HITL documentation for context.
  • S3 connector bucket listing: The S3 filesystem connector now lists only the buckets that the configured credentials are permitted to browse, avoiding access errors during connector setup.
  • Tool-run log streaming: Tool-run logs are now streamed to the workflow execution UI with markdown rendering for clearer real-time visibility.
  • Modification timestamps: Fixed modified_at timestamps not updating on several database models.

v0.160.0​

Release Date: April 23, 2026

New Features​

  • Human-in-the-Loop (HITL) audit logging and change tracking: HITL reviews now record a complete audit trail of edits, with per-field change history and review statistics for compliance and operational visibility.
  • Gemini embedding adapter: Added a dedicated Gemini embedding adapter for Google AI Studio, enabling vector embeddings with a simple API key — no GCP service account required. Defaults to the gemini-embedding-001 model.
  • Unified retry for LLMs and embeddings: LLM and embedding calls now retry transient provider failures consistently.

Bug Fixes​

  • HITL polish: Restored the change-history drawer, allowed workflow deletion when HITL audit logs are present, and fixed adapter modal layout inside the audit UI.
  • SharePoint connector: Fixed SharePoint site URL and drive ID persistence, and made OAuth token refresh safer to prevent stale-token failures.
  • Pipeline outputs: Removed a duplicate intermediate-file write during pipeline output handling.
  • MinIO cleanup: Hardened the MinIO object cleanup fallback so deletes succeed when the primary path fails.

v0.159.1​

Release Date: April 16, 2026

New Features​

  • Gemini LLM adapter: Added a dedicated Gemini adapter for Google AI Studio, enabling direct access to Google's Gemini models via an API key without requiring a GCP service account.

Bug Fixes​

  • SharePoint OAuth: Fixed multi-tenant authentication and a duplicate OAuth button on the SharePoint connector setup flow.
  • Connector forms: Connector configuration forms now pre-populate with schema defaults, reducing setup effort.
  • Dashboard polish: Hid the welcome card on the dashboard, corrected a usage count mismatch on the subscription tab, and aligned the dashboard header style with other pages.
  • Null-response handling: Improved handling of null responses from upstream services to prevent occasional errors during extraction.

v0.158.4​

Release Date: March 27, 2026

Bug Fixes​

  • Extraction playground cleanup: Removed the TSV view and download button from the extraction playground for a streamlined experience.

v0.158.2​

Release Date: March 25, 2026

Bug Fixes​

  • Security improvements: Added input validation and Content Security Policy (CSP) headers across the platform for improved protection against cross-site scripting.

v0.158.1​

Release Date: March 24, 2026

New Features​

  • AWS S3 IRSA authentication: You can now authenticate S3 storage and S3 connectors using IAM Roles for Service Accounts (IRSA), eliminating the need for static credentials in EKS deployments.
  • Usage sync failure alerts: Automated Slack notifications are now sent when usage synchronization encounters errors, improving operational visibility.
  • Improved error messages: Adapter names are now included in error messages, making it easier to identify which adapter is causing issues.

Bug Fixes​

  • Fixed TTL settings and NaN display in API HITL review configuration.
  • Improved cookie security attributes and XSS prevention headers.
  • Fixed Azure OpenAI connection errors by upgrading the underlying LLM library.
  • Fixed LLM compatibility issues with retriever-based workflows.
  • Fixed an issue where LLM refusal responses could cause unexpected errors.

v0.158.0​

Release Date: March 19, 2026

New Features​

  • Platform API keys: Added Platform API key support for programmatic access to the Unstract Platform API. Organization admins can create, list, update, rotate, and delete API keys via the UI under Platform > Platform API Keys. Keys provide Bearer token authentication for all Platform API endpoints. See Platform API Keys documentation for details.
  • Service account access bypass: Service accounts now have visibility into and can bypass access restrictions for pluggable apps, enabling automated integrations without manual permission grants.

v0.157.0​

Release Date: March 17, 2026

New Features​

  • High-availability infrastructure: Added Redis Sentinel and RabbitMQ quorum queue support for improved reliability and automatic failover in production deployments. See HA Deployment for details.
  • HITL reviewer name display: Documents in review now show the reviewer's name in the HITL queue. See Human in the Loop (HITL) documentation for details.
  • Improved HITL permissions: Streamlined permission checks for HITL operations.

Bug Fixes​

  • Fixed Redis Sentinel and MinIO HA compatibility issues across multiple services.
  • Fixed Agentic Prompt Studio configuration for on-prem deployments.

v0.156.2​

Release Date: March 16, 2026

New Features​

  • Redis Sentinel HA support: Added Redis Sentinel support for backend, Celery, cache, tool containers, sidecars, and metrics services with automatic failover. See HA Deployment for details.
  • RabbitMQ HA with quorum queues: Added configurable quorum queue support for RabbitMQ broker. See HA Deployment for details.
  • LLMWhisperer client retry backoff: Added configurable retry and backoff for LLMWhisperer client requests.

Bug Fixes​

  • Fixed Redis Sentinel stability issues across multiple backend services.
  • Fixed TTL time display to show days instead of only hours.
  • Fixed profile dropdown menu items alignment.
  • Fixed Agentic Prompt Studio build by including prompt files.

v0.156.1​

Release Date: March 11, 2026

Bug Fixes​

  • Fixed MinIO Operator Helm repo configuration in GCR push workflow.

v0.156.0​

Release Date: March 11, 2026

New Features​

  • HITL default TTL changed to 90 days: Default TTL for HITL records changed from unlimited to 90 days. See HITL documentation for details.
  • HITL Fetch Specific: Added targeted document review — fetch specific documents from the HITL queue by identifier. See HITL documentation for details.
  • HITL rule string in API responses: API responses now include rule_string and display TTL in days. See HITL Rule Configuration for details.
  • MinIO HA support: Added optional MinIO HA via MinIO Operator for high-availability object storage. See HA Deployment for details.
  • Azure AI Foundry adapter: Added Azure AI Foundry as a new LLM adapter option.
  • SharePoint/OneDrive connector: Added SharePoint and OneDrive filesystem connector support. See Working with Connectors for details.
  • Secure cookie settings: Enabled secure cookie configuration in Helm values.
  • HubSpot integration plugin: Added HubSpot contact property update integration for user journey tracking.

Bug Fixes​

  • Fixed HITL API rules still being evaluated after removal.
  • Fixed secure cookie configuration for improved session security.
  • Fixed connector directory listing cache, including Azure storage expiry regression.
  • Added PDF viewer error fallback when document fails to load.
  • Fixed landing page content clipping on laptop/tablet screens.
  • Fixed consistent empty state when search returns no results.
  • Fixed SharePoint directory listing not returning file details.
  • Improved model pricing cache refresh frequency.

v0.155.0​

Release Date: March 6, 2026

New Features​

  • Vertex AI vertex_location support: Added vertex_location field for Vertex AI adapters to configure regional endpoints. See Google / VertexAI Gemini Pro adapter for details.
  • total_pages_processed in API response: Execution API response and worker destination metadata now include total_pages_processed count. See Execution API for details.

v0.154.0​

Release Date: March 5, 2026

New Features​

  • CSV, TXT, and Excel file support in Prompt Studio: Prompt Studio file converter now supports CSV, TXT, and Excel files in addition to PDF and images. See Adding and Managing Documents for details.
  • HITL sidebar navigation and queue management: HITL moved to sidebar navigation (removed from profile dropdown), added queue deletion support and nested table row management. See Human in the Loop (HITL) for details.
  • Dashboard metrics system: Added plan banner, welcome card, and subscription usage tab to the dashboard with metrics worker Helm deployment.
  • Card-based layout for deployments: Pipelines and API Deployments listing pages redesigned with card-based layout. See API Deployments for details.
  • Profile page with role and organization info: Profile page now displays user role and organization details.
  • Sidebar expand-on-hover: Enhanced sidebar UX to expand on hover for better navigation.
  • Documentation link popover: Added documentation link popover for connectors in the UI.
  • Workflow deletion error detail: Deletion error messages now show specific pipeline/API deployment names.
  • Login page redesign: Redesigned login form with light-themed right panels.
  • User account migration: Added support for handling email changes during OAuth authentication.

Bug Fixes​

  • Fixed workflow executions getting stuck during high-load periods.
  • Fixed Vertex AI pro models incorrectly applying thinking mode when disabled.
  • Fixed Azure OpenAI cost tracking displaying incorrect model names.
  • Added beta tag to Agentic Prompt Studio navigation item.
  • Fixed logs screen pagination in card-based deployment layout.
  • Fixed Prompt Studio tool settings not preserving registry metadata on save.
  • Improved error response when a required tool image is unavailable.
  • Fixed admin access verification on navigation to admin routes.
  • Improved frontend build stability and plugin compatibility.
  • Cleaned up legacy worker infrastructure for better maintainability.

v0.153.0​

Release Date: February 26, 2026

New Features​

  • OAuth product scope: Added product scope to OAuth authorization requests for login/signup, enabling product-specific authentication flows.
  • Plugin loading infrastructure: Added dynamic plugin loading for enterprise components, cloud migrations, and rule engine.

Bug Fixes​

  • Fixed forbidden email error handling during Auth0 OAuth callback.
  • Optimized worker queries and retry configuration for worker base client.
  • Improved dev experience with compose debug override.

v0.152.1​

Release Date: February 12, 2026

Bug Fixes​

  • Added optimized HITL queue metadata backfill migration for improved performance.

v0.152.0​

Release Date: February 10, 2026

New Features​

  • HITL PostgreSQL migration: Moved HITL file handling from Redis to PostgreSQL for improved reliability and persistence. See Human in the Loop (HITL) for details.

Bug Fixes​

  • Significantly improved HITL queue performance with optimized database queries and faster queue details loading.
  • Fixed HITL TTL display incorrectly showing "Never expires" when a TTL was configured.
  • Fixed expired records still appearing in HITL queue counts.
  • Improved LLMWhisperer extraction logging for better troubleshooting.

v0.151.0​

Release Date: February 3, 2026

New Features​

  • Agentic Prompt Studio: New backend and frontend for agentic document extraction workflows (beta). Includes SDK integration, registry support, and dedicated navigation. See Agentic Prompt Studio documentation for details.
  • Profile page redesign: Added role and organization info display with enterprise design.
  • Copy-to-clipboard for file execution ID: Added copy button for execution ID in LogModal.

Bug Fixes​

  • Fixed HITL rule engine nested array flattening for multi-COA documents.
  • Fixed highlight offset issue.
  • Fixed SIGTERM trap handlers for graceful container shutdown.
  • Fixed incorrect constant reference breaking cloud config loading.
  • Improved save UX in Prompt Studio settings components.
  • Fixed dashboard icon always appearing highlighted.
  • Removed success alerts from deployments and pipelines.

v0.150.1​

Release Date: January 30, 2026

Bug Fixes​

  • Fixed incorrect constant reference breaking cloud config loading.

v0.150.0​

Release Date: January 22, 2026

New Features​

  • 1M context for Anthropic LLM adapters: Added support for 1M context window for Anthropic LLM adapters. See Anthropic LLM adapter for details.
  • Billing portal rename: Removed "Past Invoices" button and renamed to "Manage Billing" in subscription UI.

Bug Fixes​

  • Fixed custom data string values being wrapped in extra quotes. See Custom Data Variables for details.
  • Added custom data support to single pass extraction. See Custom Data Variables for details.
  • Fixed PostgreSQL race condition in concurrent table creation.
  • Fixed memory leak by closing database cursors in subscription usage handler.
  • Fixed resource leaks in platform-service.
  • Added improved logging for retrieval operations.

v0.149.0​

Release Date: January 20, 2026

New Features​

  • Stripe billing portal: Added Stripe billing portal integration for LLMWhisperer subscription page.
  • Usage billing control: Added ability to disable usage billing for specific subscriptions in API Hub.

Bug Fixes​

  • Fixed packet processing final fetch response API.
  • Added database fallback for API key lookup in LLMWhisperer Bridge.
  • Fixed role update failures for existing users.
  • Fixed Ollama adapters failing post LiteLLM migration. See Ollama adapter for details.
  • Fixed export reminder state persistence across page reloads.