Crogl 2.0.0 Release Notes

We're excited to announce the release of Crogl 2.0.0. This major release puts your work front and center with new role-based dashboards, adds scheduled jobs and richer investigation reports, and makes the agent markedly smarter with a rewritten knowledge graph. Under the hood, 2.0.0 re-architects the agent runtime by migrating to OpenCode, and gives every conversation its own isolated container, alongside a round of connector, model, installer, and security improvements. We encourage all customers to upgrade.

Note on versioning: With 2.0.0, Crogl moves to a standard major.minor.patch release scheme. Version numbers going forward follow that convention.

What's New

• Dashboards
• Rewritten Knowledge Graph
• Scheduled Jobs
• Richer Investigation Reports
• Re-architected Agent Runtime (OpenCode)
• Per-Session Container Isolation
• Configurable Model Context Windows
• Expanded and Redesigned Model Configuration
• Community Connectors
• Improved Connector Tool Discovery
• Resilient Connector Schema Refresh
• Hardened Installer and Network Defaults
• Security Hardening

Detailed Changes

Dashboards

Crogl now opens on role-based dashboards that surface the work that matters to you. Analyst views track your open tickets, what you have analyzed, and your resolution time, while the admin view rolls the same signals up across the team, including a Crogl Adoption card that shows how much of your ticket volume the team has investigated. The team views count your whole team's work rather than every user in the connected ticketing system, and the Adoption card credits an investigated ticket by matching the ticket itself, so it counts your team's work even when the ticketing system reports assignees only as display names.

Rewritten Knowledge Graph

Crogl now builds a knowledge graph from your connected data sources, capturing relationships between entities so the agent can answer broad questions that span multiple connectors. Instead of treating each source in isolation, the agent can follow an entity across your stack and reason about how findings connect.

Scheduled Jobs

You can now put recurring work on a schedule. The Crogl assistant can set up and manage recurring scheduled jobs from chat, and it can schedule a recurring job that continues in your current conversation, appending each run to the same thread instead of starting a fresh one. New crogl CLI commands create, list, and manage scheduled jobs, and the run lists support pagination, filtering, and sorting so long histories load a page at a time.

Richer Investigation Reports

Investigation reports can now display an interactive map of how the agents communicated during an investigation, giving you a clearer picture of the reasoning behind a conclusion. Richer Markdown formatting, including tables, checklists, and strikethrough, now renders consistently across chat, investigations, and the agent view.

Re-architected Agent Runtime (OpenCode)

The agent runtime has been re-architected on OpenCode. This migration modernizes the foundation the agent runs on and sets up the reliability and performance work that follows in later releases.

Per-Session Container Isolation

Each conversation now runs in its own fully isolated container, improving separation between concurrent sessions and keeping one session's work from affecting another.

Configurable Model Context Windows

You can now set the context window and maximum output length per model. Supported models pre-fill their recommended limits and let you raise them up to the model's published maximum, while custom or self-hosted models let you enter the values freely. This keeps the agent's history management aligned with the real model. A configured context window or output length above the model's maximum is now rejected, and each supported model carries both a recommended context window and its published maximum.

Expanded and Redesigned Model Configuration

The model-settings experience in onboarding and in Settings > Models has been redesigned around a connected-model status card with Edit, Change Model, and Test connection actions, plus a guided flow for connecting or switching models. Recommended models are grouped by brand (Anthropic, OpenAI, Google, Mistral), the picker shows only models Crogl has validated, and you can still connect custom providers, endpoints, and on-prem models, now with an option to skip TLS verification.

The supported-model catalog expands to include Claude Opus 4.8, Claude Fable 5, Sonnet 5, Opus 4.7, and Haiku 4.5; GPT-5.1, GPT-5.2, and GPT-5.3 Codex; Google Gemma 4 31B; and Mistral Devstral 2, across Anthropic, AWS Bedrock, OpenAI, and self-hosted providers. Language-model endpoints served over TLS 1.2 are now supported.

Community Connectors

Administrators can import community-developed connector types from a bundle file and configure them like built-in connectors. Imported types appear in the catalog marked "community." An administrator can also delete an imported community connector type. By default the request is blocked while instances are configured, and an explicit cascade option removes those instances first. Built-in connector types cannot be deleted. Creating a connector now survives a brief network interruption, such as a VPN switch, a Wi-Fi or Ethernet handoff, or waking from sleep. The request retries automatically, and the connector is never left partially set up if the connection drops mid-save.

Improved Connector Tool Discovery

Crogl improves how the agent discovers the tools available for each connector, and it handles lookups that span multiple connectors more effectively.

Resilient Connector Schema Refresh

Connector searches for Cribl, Splunk, Databricks, and Elasticsearch, along with the scheduled schema refresh, are now more resilient to slow data sources. A slow response is retried instead of failing the whole operation, and one slow part of a refresh no longer aborts the rest. Schema refresh no longer drops valid fields that did not appear in a given sample. Splunk searches now release their server-side search jobs as soon as results are retrieved, preventing a buildup of artifacts that could exhaust Splunk's disk quota and stall new searches.

Hardened Installer and Network Defaults

Self-hosted installations now bind the Crogl UI, API, and bundled database to localhost by default, reducing network exposure. After setup, the installer prints a summary of the network ports each service uses, with guidance on which to expose. Installers offer a clean-install option when they detect an existing installation, and a new crogl.sh restart command stops and starts an installation in one step. Linux direct-access installs can specify an explicit IP address with --ip for the browser endpoint and TLS certificate, and they reliably preserve the detected hostname. Installer reruns no longer prompt for certificate-trust approval when the existing certificate is reused, and certificate generation no longer fails when system or user OpenSSL configuration is present.

Security Hardening

Crogl adds sign-in through Microsoft Entra ID, alongside other standard OIDC identity providers. Sign-in security is strengthened, and standard browser security protections now apply to all web responses.

Bug Fixes

• Fixed intermittent sign-in failures ("token is not valid yet") that could occur when your identity provider's clock differed slightly from Crogl's.
• Fixed Docker startup failures on RHEL-family 10 systems by checking required networking kernel modules first.
• Fixed a permission error that could prevent the bundled PostgreSQL from starting on some Linux hosts.
• Fixed intermittent "Could not get lock /var/lib/dpkg/lock-frontend" failures on fresh Ubuntu hosts; the Linux installer now waits for the system package lock and suppresses automatic service restarts while installing prerequisites.
• Fixed errors when using newer GPT models, including GPT-5.4 and GPT-5.5.
• Fixed Save and Test on language-model settings so it no longer intermittently validates against the previous provider; the connection test now checks the configuration you just saved.
• Fixed the Crogl Assistant panel in investigations so it scrolls to the bottom; the latest assistant message is no longer cut off. The panel now shows the assistant's full activity, including the tools it runs, their results, and inline hints, laid out compactly to fit the narrower panel.
• Fixed a case where a chat response could be lost if its stream connection dropped near the end of a long turn under heavy load; reconnecting now replays the completed turn from the saved record instead of failing.
• Fixed a bug where a chat response could appear twice immediately after it finished streaming; the reply now renders exactly once.
• Fixed a case where the assistant could appear to freeze when it wanted to ask a clarifying question.
• Fixed a server crash that could occur when the upstream language model dropped a streaming connection mid-response; the request now fails gracefully instead of taking down the server.
• Fixed the admin dashboard's Crogl Adoption card, and the "analyzed" figures on the Open Tickets and Resolution Time cards, so they count your team's investigations rather than reading zero.
• Fixed the Crogl Adoption card's "All users" view so it loads reliably when a Crogl login has no matching user in the connected ticketing system.
• Fixed your dashboard's Open Tickets card so its "Analyzed" list shows only investigations you ran, and added a clear message with a Retry button when an investigation cannot be opened.
• The published REST API reference now accurately reflects crogld's current endpoints and is generated from the server code so it stays in sync going forward.