Alpha. Under active development. Expect rough edges.
orqi (aka TonyBot) is the orq.ai helper agent in your terminal.
Ask about your workspace in plain words. It answers with your workspace's own tools, models and skills, already connected.
A session on orq/auto. Each message is routed to a model, and the sidebar shows the trajectory, the context, memory and what routing saved.
-
Install
curl -fsSL https://raw.githubusercontent.com/orq-ai/orqi/main/install.sh | sh -
Run
orqi -
Sign in when it asks: in the browser, or paste an API key.
-
Type
/learnfor six guided first prompts.
Keep the orq CLI on your PATH. orqi uses it to sign in and to
find your workspace. More install options: docs/install.md.
- "Why is my agent failing?"
- "Which model costs the most this week?"
- "Build an evaluator for hallucinations."
- "Compare these two agents."
- "How do I invoke my deployment from Python?"
| Command | What happens |
|---|---|
orqi |
Opens the app |
orqi "why did my agent fail today?" |
Answers once, then exits |
orqi --session <id> |
Reopens a past session. The id is printed when one ends |
- Already connected. The workspace's models, orq.ai tools and skills load at start.
- You see what it does. The sidebar shows every step, warnings, context use, routing and memory.
- Cheaper answers.
orq/autosends easy messages to a fast model and hard ones to a strong one. - Safe by default. Deletes and other calls that cannot be undone ask first. So do shell commands.
- Plan first.
/planmakes the session read-only until you approve a plan. - Talk to your agents.
/agentslists the workspace's agents, and/agent <key>talks to one.
Harness
| Command | Does |
|---|---|
/plan |
Read-only mode. Look, ask, plan. Nothing changes until you approve |
/handoff <goal> |
Starts a fresh session with a drafted summary of this one |
/route |
How orq/auto picks a model, and its settings |
/context |
What fills the context window |
/cost |
What this session cost, per model |
/btw <question> |
A side question on the cheapest model, kept out of the chat |
alt+s |
Shows or hides the sidebar |
orq.ai platform
| Command | Does |
|---|---|
/agents |
The workspace's agents: talk, inspect, traces |
/trace |
The last answer's trace: route, spans, cost |
/overview |
Agents, prompts, datasets, experiments, evaluators |
/garden |
The model garden in your browser |
/skills |
The loaded skills |
/learn |
Guided prompts, one per platform area |
/whatsnew |
The latest orq.ai release |
Keys: esc interrupts · ctrl+c exits · / lists every command · ! runs a shell command ·
ctrl+o expands a tool result
Every command and key: docs/commands.md.
Run orqi update. orqi tells you when a release is out. It never updates itself.
- Not signed in?
/loginor/key. - Wrong workspace?
/workspacepicks another. - Something broken?
/doctorchecks your setup. - Found a bug?
/bugopens a prefilled issue.
Which credential wins and what each sign-in error means: docs/credentials.md.
| Doc | For |
|---|---|
| docs/commands.md | Every command, key and safety prompt |
| docs/configuration.md | Environment variables, themes, where settings live |
| docs/install.md | Installers, platforms, checksums, building from source |
| docs/credentials.md | Credentials, hosts, profiles, sign-in errors |
| ARCHITECTURE.md | How it is built |
| AGENTS.md | Rules for changing the code, and releasing |
| SECURITY.md | What orqi downloads and runs |
MIT licensed. Built by orq.ai research labs.