Changelog

v0.14

• Added Slurm Linux user binding: users can bind a Linux account from Profile, and project owners can choose to run supported project work as the bound identity
• Added Codex OAuth credential management in model forms: log in to Codex, import a local Codex login, refresh tokens, and clear saved credentials for the OpenAI Codex template
• Enhanced file preview: HTML files and Visualizer plugins can open in the preview modal for custom project visualization pages
• Added conversation task progress panels for multi-step assistant work
• Added richer real-time tool output streaming and safer stop behavior for long-running turns
• Improved context compaction for long tool-heavy turns by allowing mid-turn checkpoints between tool calls
• Improved chat file references so workspace paths, bare paths, and file:// links are easier to recognize and copy
• Added provider templates for OpenAI Responses, OpenAI Codex, OpenRouter, DeepSeek, and related model providers, with a unified 200k-token default context window

v0.13

• Added Assistant saved configs and template management: projects can save, update, apply, share, and delete config snapshots; admins can review, pin, order, and maintain recommended templates
• Added conversation-level Assistant use, which affects only the current conversation without rewriting project defaults
• Improved conversation branching: editing historical user messages creates message branches and syncs child conversation events for easier prompt comparison
• Improved chat messages, file attachments, and composer insert actions; bash, write, and edit tool messages now use clearer collapsible sections with smarter default-open behavior
• Enhanced sidebar workflows: cross-project new chat, project creation from the project list, and show more/less pagination for project conversations
• Added saved user preferences for language, theme, and interface mode across pages
• Added Skill installation directly from catalogs
• Project creation can optionally initialize the first conversation, and project settings support project deletion
• Docker and Kubernetes persistent containers support restart; Kubernetes mode adds startup diagnostics, registry prefix handling, workspace-backed asset repositories, and Kaniko build backend support
• Project, user, memory, and running-container lists now use server-side pagination
• Saved config import/export now has a PowerShell zip fallback on Windows

v0.12

• Added Retrieval Memory (mem0) — admins can enable globally with embedder configuration (OpenAI / Ollama); each chat turn auto-searches historical memories and async-writes new memories; projects can toggle independently
• Fixed Docker Compose retrieval-memory persistence so mem0 no longer defaults to the image-internal /app/data/mem0 path in server deployments; mem0 now persists under ${DATA_DIR}/mem0 on the host and maps to /data/mem0 in the container
• Added Resource Quota Management — admins can set platform-level and project-level CPU, memory, GPU, and container instance limits to prevent resource contention
• Added Execution Platform Settings — admins can configure the container orchestration backend mode (Docker currently supported)
• Added Container State Sync — after host restart, container states and resource leases are automatically reconciled
• Added Resource Wake — waiting containers and Agent tasks are auto-retried when resources become available
• Added Admin Navigation Grouping — admin sidebar organized into collapsible categories (Operations, Capabilities, Security, Infrastructure)
• Enhanced Showcase — public pages now render Agent task cards, visualizer plugins, and file tree browsing
• Added OpenAI-compatible model thinking config — supports configuring thinking parameters for openai-compatible models to enable extended reasoning
• Enhanced Agent output collection — improved file references, glob patterns, content reading, and output expression capabilities
• Enhanced Asset Management — admin page supports Visualizer plugin import/export and inline editing

v0.11

• Added public Showcase page — admins can publish projects to /showcase for visitors to browse conversations and file previews
• Added local workspace cross-project file import — select source project, pick files/directories, and import into current target directory
• Chat input supports @ file search to directly attach project files
• Long conversations now have a message anchor navigation bar for quickly jumping to user inputs and compaction nodes
• Registration, profile edits, and member invitations now validate email format; project invitations only accept registered emails

v0.10

• Added personal model management at /workspace/models and scoped model access for chat/model pickers
• Enforced role-based access for project workspace writes and project container start/stop operations

v0.9

• Added Token Usage dashboard — personal model token consumption for the last 30 days, grouped by day, month, project, model, provider, or source
• Added Action batch selection, zip export/import, and cross-project import
• Web search tool is now gated behind admin feature flag; agents cannot call web_search when disabled

v0.8

• Added WDL-style input/output definition support for Agent Actions
• Standardized Agent run output directories to agents-output/<YYYY-MM-DD>/<agent-name>-<ticket-id>
• Upgraded Agent Workflow result definition mode (result.outputs replaces result.step_id)

v0.7

• Added Persistent Container Execution Without runtime_ref documentation — Process Actions can run in project persistent containers without explicit runtime_ref
• Added Container CPU and Memory Resource Settings documentation
• Added Model provider_spec, policy, and capabilities configuration documentation

v0.6

• Added Container Fixed Mounts documentation — admin-managed bind mounts auto-injected into persistent and agent containers
• Updated Collaboration and Permission Boundaries with admin container mount management role

v0.5

• Added Prompt Action editor feature documentation
• Added Agent ticket cancellation feature documentation
• Added project mode awareness feature documentation

v0.4

• Added Skill and Agent import/export feature documentation
• Added conversation compaction state visualization documentation
• Added Electron desktop app support documentation
• Added smart login redirect feature documentation

v0.3

• Skill installation now uses namespaced layout .agent/skills/<namespace>/<skill-name> to avoid collisions
• Legacy flat installs .agent/skills/<skill-name> remain compatible
• agent-authoring skill in dev mode is exposed from the read-only global skill mount
• Improved error messages when skill name conflicts occur

v0.2

• Split manual into multi-page framework
• Aligned structure with vision, personas, and MVP scope
• Improved left sidebar navigation readability

v0.1

• Initial single-page manual