Totem

Capabilities

What Totem covers, grouped, each with the phase that delivers it (roadmap). The list is the scope of a complete coding agent; nothing in it depends on a particular existing product.

AreaCapabilityPhase
ModelsProviders by parser: OpenAI-compatible (includes OpenRouter), OpenAI Responses, Anthropic, Gemini, Ollama1 (one), 3 (all)
Model list per provider, default model, per-session model1 / 3
Bundled catalogue of free and local models, editable by hand3
Reasoning/think mode where the provider supports it3
Token and cost counters per step and per session3
Retry with backoff, context compaction2
AgentLoop with streaming, abort, step limit1
Tools: read, glob, grep, write, edit, bash1
Tools: patch, fetch, todo, ask, skill2
Sub-agents and agent definitions in files3
Instructions from AGENTS.md (project and nested)2
Skills (folders with SKILL.md)2
Custom commands (/name from files)3
MCP client (stdio, HTTP)3
MCP server (selected actions via /$mcp/)3
Scheduled prompts5
ProjectsOpen a folder, detect Total.js projects, list sessions1
File tree and viewer1
Git awareness: branch, status, working tree diff2
Worktrees for parallel sessions5
Attachments: images and files in the prompt3
ReviewPer-session diff, undo per message (snapshots)2
Split diff, accept/reject per file3
SafetyPermission rules with ask/allow/deny, project boundary1
Auto-accept edits mode2
Output secret scanning2
InterfaceChat, parts rendering, projects, settings1
Terminal output panel2
Interactive terminal (PTY)4
Language server diagnostics fed to the agent4
Shortcuts, themes, translations2
Deliverytotem command, totem run headless2
Docker image5
Native windows (macOS, Windows, Linux), signed6
Total.jsA skill pack and instructions for Total.js 5 projects (the conventions plugin)2
Knowledge tools: search the Forge documentation from the agent3

Not planned#

Hosted accounts, usage billing, a built-in model, voice input, sharing sessions through a public service. They can be revisited, each as its own decision.

Edit this page on GitHub