Multi-Provider Model Support
OpenAI GPT-5 & GPT-4
- GPT-5.2 Flagship
- GPT-5.1 / GPT-5 Mini / Nano
- GPT-5.2 Codex Code
- GPT-4o / 4o-mini
- GPT-4o + Search Search
- o1 / o3-mini Reasoning
Anthropic Claude 4
- Claude Opus 4.5 Most Capable
- Claude Sonnet 4.5 Balanced
- Claude Haiku 4.5 Fast
- Claude Sonnet 4
- Extended & Adaptive Thinking: Opus 4.5, Sonnet 4.5
- Code Execution: 4.5+ family
Google Gemini 2.5
- Gemini 2.5 Pro Frontier
- Gemini 2.5 Flash Fast
- Gemini 2.5 Flash Lite
- Gemini 1.5 Pro (Legacy)
- Gemini 1.5 Flash (Legacy)
- All via OpenAI-compat proxy
Extended & Adaptive Thinking
Off
Standard response, lowest latency, all models
Adaptive — Fast
Claude 4.6 family, ~2K token budget, simple tasks
Adaptive — Balanced
Claude 4.6 adaptive, ~5K budget, mixed workloads
Deep Think
Claude 4.5/4.6, up to 10K tokens, complex reasoning
Model-aware toggle automatically applies adaptive (4.6) or extended (4.5) API parameters — Reasoning blocks collapsible inline
Core Chat Features
Prompt Library Built-in
Curated and community-built prompts organized by category with AI scoring and guided builder.
- Browse tab: curated + community prompts, search & category filter
- Workbench builder: system + user panel, variable detection
- AI scoring via Claude: reliability, efficiency, clarity scores
- Submit tab: POST to MongoDB with tagging
- Use in Chat: sets system prompt separately, fills user input
Agent Skills New
Claude and OpenAI agent skills — reusable SKILL.md bundles that provide domain-specific capabilities activated per conversation.
- Browse pre-seeded skills: Financial Analyst, Code Review, SQL Optimizer
- SKILL.md builder with YAML frontmatter and markdown body
- Provider-tagged (Claude / OpenAI) with version management
- Activate to inject skill context as system prompt
- Create and save custom skills to team library
Team Channels Collaboration
Persistent team workspaces for shared AI conversations organized by function or project.
- Named channels visible to all team members
- Conversation history per channel via MongoDB
- Channel context preserved across sessions
- Toggle between personal and team views
Code Execution Tools Claude 4.5+
Inline code execution via Anthropic's sandboxed environment with programmatic tool calling support.
- Toggle Code pill enables code_execution_20260120 tool
- Output rendered inline as terminal-style blocks
- Stdout, stderr, exit code displayed per execution
- Claude can call tools programmatically from code
File Attachments & Vision
Multi-modal inputs: images, PDFs, documents, code files across all providers that support them.
- Drag-and-drop or click to attach images, PDFs, docs
- Provider-aware: native PDF for Claude, text extraction fallback
- Image inline preview before send
- Document links for non-image file types
Custom System Prompts
Active system prompt indicator visible in chat header — set via Prompt Library, Agent Skills, or per-assistant presets.
- 6 built-in assistant personas (Compliance, Code, Analyst, etc.)
- Purple pill indicator when custom system prompt is active
- One-click clear with session context preservation
- Prompt Library "Use in Chat" sets system + user separately
Cache & Cost Intelligence
Per-Message Indicators
- Cache Hit / Miss badge on every response
- Similarity percentage for semantic cache hits
- Tokens saved counter per message
- Session-level stats in header bar
LLM-Side Prompt Caching
- Auto-inject Anthropic cache_control headers
- Large system prompts (>4K chars) always cached
- Repeated medium prompts cached after threshold
- OpenAI automatic caching tracked passively
Session Cost Tracking
- Running input/output token counters
- Estimated session cost (USD)
- Water bottle equivalent CO2 metric
- Cache savings in token and cost terms
Capability Matrix by Provider
| Capability |
OpenAI GPT-5 |
Anthropic Claude 4.5 |
Anthropic Claude 4.6 |
Google Gemini 2.5 |
| Standard Chat |
Yes |
Yes |
Yes |
Yes |
| Extended Thinking |
— |
Yes (extended) |
Yes (adaptive) |
— |
| Code Execution Tool |
— |
Yes |
Yes |
— |
| Vision / Image Input |
Yes (GPT-4o+) |
Yes |
Yes |
Yes |
| PDF Native Support |
Text extract |
Native PDF |
Native PDF |
Text extract |
| LLM-Side Prompt Cache |
Auto (OpenAI) |
Injected |
Injected |
— |
| Agent Skills (SKILL.md) |
Yes |
Yes |
Yes |
— |
| Prompt Library |
Yes |
Yes |
Yes |
Yes |
| AI Prompt Scoring |
— |
Via Claude proxy |
Via Claude proxy |
— |
Deployment & Identity
Authentication & SSO
Integrates with Smartflow's unified identity layer for transparent corporate authentication.
- LDAP / Active Directory / Azure AD authentication
- OIDC / SAML SSO via trusted proxy headers
- JWT session propagation to all proxy requests
- Group-based AI policy enforcement via Smartflow
Smartflow Proxy Integration
All AI requests route through the Smartflow governance proxy for policy, caching, and logging.
- Compliance policy enforcement per user/group
- 3-tier MetaCache (memory + Redis + MongoDB)
- VAS logging for full audit trail
- Prompt compression & compression savings tracking