AI-assisted coding
without leaving the shell.
VeloxCLI is an open-source TUI and CLI for serious developers. Inference defaults to Groq—most turns finish in under two seconds, with throughput and pricing tuned for tight feedback loops.
Transparent memory via /smart-dream and Kairos, hardened workspace permissions, and /resume for tmux and SSH are first-class—not afterthoughts.
Under 2 seconds typical end-to-end latency on Groq in normal conditions. You bring the API key. VeloxCLI is independent of Groq, Inc.
A fast, inspectable terminal assistant
VeloxCLI pairs a modern TUI with a scriptable CLI. Groq handles inference so you get speed and cost structure suited to high-frequency coding—not a bloated browser tab or a subscription wrapper around someone else's model. Memory consolidation is designed to be understandable: you can see what Smart Dream does instead of guessing what the assistant remembers.
- SPEED
Groq-first path targets sub-two-second responses and high tokens-per-dollar for the edit-compile-test loop. Slow assistants train you to avoid asking questions; fast ones change how often you reach for help.
- MEMORY
/kairos and /smart-dream implement transparent consolidation: context is folded with intent you can trace, not a hidden summary you cannot audit.
- EXEC
/run executes shell tools with confirmation;/watch ties the session to filesystem signals. Logs stay in the transcript for review.
- TRUST
Workspace-locked file access, unified permission checks, trusted hooks, and no raw API keys in logs—shaped by a recent security audit. Privacy defaults and Groq data handling are documented; you stay in control of what leaves the box.
How it looks in the terminal
One surface: status, commands, and recent work. No fake window chrome, no stock photos—just the layout your ssh session would show. Replace this block with a screen recording or GIF when you ship marketing assets; the structure below matches the real TUI.
/resume ready for tmux / SSH
kairos memory mode
/smart-dream consolidate context
/help slash index
/run cargo test (approved)
Attached src/ via /files
mcp attach internal API
session export markdown
Built for velocity, clarity, and production shells
Groq speed and economics
Default inference on Groq keeps median turns under about two seconds in typical use. High throughput and straightforward per-token economics matter when you iterate all day—not only on demos.
Smart Dream and Kairos
Transparent memory consolidation: Smart Dream folds context deliberately; Kairos holds assistant-mode state. You are not stuck with an opaque summary you cannot reason about.
Exec mode that respects the shell
/run for tools with explicit approval, /watch for file signals. Output stays in the thread—useful for postmortems and code review.
Security you can explain
Workspace locks, unified permissions, trusted hooks, and disciplined secret handling follow a recent audit. The goal is fewer foot-guns than ad-hoc agent scripts.
/resume and terminal resilience
Detach, reconnect, tmux, SSH: /resume is aimed at real remote workflows, not only local laptops.
/focus and automation
/focus trims noise for sustained speed.--print and JSONL-friendly output fit CI, scripts, and pipelines—same binary as the TUI.
Why VeloxCLI instead of another terminal agent
Other tools overlap on “AI in the terminal.” VeloxCLI optimizes for Groq latency, inspectable memory, explicit execution policy, and automation-friendly output. The table is directional—verify against each project's current docs before you standardize.
| Dimension | VeloxCLI | Typical browser-first agents | Other terminal agents |
|---|---|---|---|
| Default inference latency | Groq-first; sub-2s typical turns, high throughput | Network and model dependent; often higher round-trip | Varies widely by provider and transport |
| Memory transparency | /smart-dream, Kairos, inspectable consolidation | Often opaque conversation history | Varies; many rely on implicit context only |
| Security posture | Audit-informed workspace lock, unified permissions, trusted hooks | Depends on host; wider attack surface in browser stacks | Implementation-specific |
| CI and scripting | --print, JSONL-oriented workflows | Poor fit for headless pipelines | Mixed; not always first-class |
| SSH / tmux resilience | /resume, terminal-native lifecycle | Usually disconnected from remote shells | Varies |
| License and client cost | MIT client; pay inference only | Often proprietary or bundled SaaS | Mixed open and commercial |
Three commands, then you are in
Pick one install path. Then export GROQ_API_KEY and run velox. Full steps and flags live on the Install page.
curl
curl -fsSL https://example.com/install.sh | sh # replace with shipped installer URL
Homebrew
brew install veloxcli
npm
npm install -g veloxcli
Audit-informed, privacy-first defaults
A recent security review hardened how VeloxCLI touches the filesystem, executes hooks, and handles credentials. Workspace-locked paths, unified permission checks, and trusted hook lists reduce the chance of a clever prompt turning into an unbounded shell. API keys belong in the environment or secure config—not in pasted logs.
Groq receives prompts and context you send for inference under their terms. The client does not add product telemetry on top. Read Privacy, the repository SECURITY.md, and the audit notes in-tree for the authoritative story.
MIT on GitHub—patches welcome
Source, issues, and discussions live on GitHub. Fork it, run it in your team, open a PR for a slash command or a plugin. Commercial use is allowed under the license; attribution is in the LICENSE file.
Common questions
Answers below track the same positioning as the rest of the site: Groq-first latency, transparent memory, and audit-informed defaults. For behavior of a specific release, use the docs in the repository.
Stop copying stack traces into a browser
Install VeloxCLI, set your Groq key, run one session with /smart-dream and /run. If the latency and transparency are not better than what you use today, you have lost one coffee—not a procurement cycle.