Skip to content

Chat

The chat workspace is the primary interface to your local Keprix agent. It supports multi-turn conversations, tool use, model switching, and slash commands.

Open chat

  • Route: /chat
  • Launcher card: Chat
  • Sidebar: Workspace > Chat

Starting a New conversation creates a session immediately; the UI does not redirect you into an old thread.

Layout

Region Purpose
Session list Past conversations, rename, delete, star
Message feed User and agent messages, tool calls, code blocks
Input bar Text input, model selector, attachments; web voice input (mic, prompts 188-191) planned
Header nav Home (/launcher) and Dashboard (/dashboard, admins only)

Models

Use the model selector in the chat header. Models come from configured LLM providers (see LLM providers).

  • Built-in: DeepSeek, Anthropic, OpenAI, Gemini, Ollama
  • Custom: OpenAI-compatible endpoints added in Dashboard > Settings > LLM Providers

Default provider is marked in admin settings.

Agent capabilities

During a turn the agent may:

  • Call registered tools (filesystem, web, workspace APIs, MCP tools)
  • Stream partial responses and thinking blocks
  • Propose mutations (new tools/skills) for owner approval
  • Run slash commands (see below)

Tool iteration limits and context compression are configured under Dashboard > Settings > Agent behaviour.

Slash commands

Type / in the input bar to open the slash menu. Examples:

Command Purpose
/opportunity Opportunity engine (see Opportunity engine)
/research Start or continue deep research
/help List available commands

Exact commands depend on installed skills and packs.

Sessions API

Action Endpoint
List sessions GET /api/conversations
Create session POST /api/conversations
Send message POST /api/conversations/{id}/messages (streaming)
Export session GET /api/workspace/sessions/{id}/export

Full paths: API reference.

OpenAI-compatible access

External clients can use POST /v1/chat/completions with a developer API key. See Developer platform and OpenAI-compatible API.

Troubleshooting

Symptom Check
No models in selector Provider keys in .env or dashboard LLM settings
Tool calls fail MCP server status, network egress, Scout kill switch
Stream stalls Backend logs, KEPRIX_MAX_TOOL_ITERATIONS
Cannot start blank chat Clear cache; ensure latest frontend (no auto-redirect to first session)