Tempr CLI · Beta

Your agent, in the terminal.

The same bring-your-own-key agent you use in your IDE, now on the command line. Fire a one-shot prompt at your repo, drop into an interactive REPL, or pipe it into a script — with the same models, personas, and provider keys as Tempr Chat.

Get started Read the docs

1 licenseIDE, terminal & API — one account

3 ways to runone-shot, REPL & piped

BYOKsame provider allowlist as Chat

tempr

~ $ tempr "add input validation to PaymentRequest.cs"
● read_file PaymentRequest.cs
● edit_file PaymentRequest.cs
● run_command dotnet build
✔ validation added — build green
~ $ tempr
> @debugger why does CheckoutTests fail █

Everything the IDE agent does, where you already work

A real tool-calling loop — not a chat wrapper. It reads and edits files, runs commands, and semantic-searches your codebase, all from the shell.

One-shot agent turns

tempr "fix the failing test in CartService" runs a full agent turn with real tool-calling — file read/write/edit, run_command, and semantic search — then hands the terminal back to you.

A real REPL

Run tempr bare for an interactive session with persistent cross-session history, proper line editing on every OS, and tab-completion for /commands and @personas.

Pipes & headless mode

Feed it stdin and read stdout: { echo "what's failing?"; cat error.log; } | tempr works like any other Unix tool, so you can script it into builds, cron jobs, and pipelines.

Specialist personas

The same named personas as the IDE: tempr "@debugger why does this fail" scopes the prompt and tool set to a specialist. @test, @review, @security, and the rest of the lineup work too.

MCP in the terminal

tempr mcp add connects the same Model Context Protocol servers you use in the IDE, so your custom tools and data sources follow you to the shell.

Same BYOK providers

Your key, your model — the full supported-provider allowlist (OpenAI, Anthropic, Google, OpenRouter, and more) works exactly as it does in Tempr Chat.

Runs on your existing license

No separate subscription. Sign in with the license key you already use — Standard, Pro, Teams, or a seat that comes with Gateway Pro or Scale — and the CLI is another surface on the same account.

Cross-platform & resilient

Self-contained builds for Windows, macOS, and Linux, on x64 and Arm64 — no runtime to install. The agent loop runs server-side like the IDE, so a flaky connection doesn't kill a long turn.

See it in action

The CLI itself, and what a terminal run leaves behind in the Portal — every agent turn is traced, costed, and searchable after the fact.

A terminal running tempr with a request to fill in an empty xUnit test: the agent reads the test, service and request files, then proposes a diff that writes the test and asks Approve this action? with Yes, Yes and trust for this session, No, and No and tell the agent why — Ask for a change in plain words. The agent reads the code it needs, shows the edit as a diff, and waits for your OK before it writes anything.

A terminal piping git diff into tempr with a request for a short review; the reply lists missing amount validation in a new RefundAsync method with a suggested fix, untracked partial refunds, and missing tests — Pipe anything in. Here `git diff` goes in, and a review of the change comes back.

A terminal running tempr keys add anthropic with the key typed at a hidden prompt and checked with Anthropic, then tempr models listing thirteen Anthropic models with their prices and context windows, then tempr config set-model choosing anthropic/claude-sonnet-5-5 — Bring your own provider key: it's typed at a hidden prompt and checked before it's saved. `tempr models` lists what it unlocks, with prices.

The Agent traces page in the Tempr Portal, listing agent runs with their persona, model, status, duration, tool-call count and cost — Every run the CLI makes lands in the Portal: persona, model, duration, tool calls, and what it cost.

The Portal's view of one agent run: alternating model calls with their token counts and cost, and tool calls such as read_file, search_files, write_file, run_tests and run_command — Open a run to see the whole loop: every model call with its tokens and cost, and every tool call it made.

Real tempr sessions on a demo project, and the runs they left in the Portal.

Capabilities at a glance

Everything the CLI can do, and the exact invocation for each — the same agent, models, and personas as the IDE.

Capability	How you use it
One-shot agent turn with full tool-calling	`tempr "fix the failing test in CartService"`
Interactive REPL — persistent history, tab-completion	`tempr`, then `/help`
Headless / piped mode	`{ echo "review this:"; git diff; } \| tempr`
Plain chat, no tools	`tempr --ask "…"`
Specialist personas	`tempr "@debugger why does this fail"`
Per-run model override	`tempr -m <id> "…"`
Run unattended: edits and allow-listed commands without asking	`tempr -y "…"`
Resume a saved session	`tempr --id <SESSION_ID> "continue"`, or `tempr -c` for this folder's last one
See and undo what the last turn changed	`/diff`, `/undo` in the REPL
Machine-readable event stream	`tempr --json "…"` — one event per line, then a result object
Token usage and cost per turn	`tempr --verbose`
Local cost by model	`tempr usage`
MCP servers	`tempr mcp add` / `tempr mcp browse` / `tempr mcp list`
Files, whitelisted commands, semantic search, symbols, tests, diagnostics	Built into every agent turn
Sub-agent delegation	Automatic — the agent hands sub-tasks to personas mid-turn
Cross-platform	Self-contained builds — Windows, macOS, Linux, on x64 and Arm64

All of it on your existing Chat license, or a seat on your Gateway plan — no separate subscription. Full command details in the command reference.

Up and running in a minute

1. Install

One line, nothing needed first — the download is self-contained.

macOS and Linux

curl -fsSL https://temprhq.io/cli/install.sh | sh

Windows (PowerShell)

irm https://temprhq.io/cli/install.ps1 | iex

Or download a build directly, or use npm, Homebrew, Scoop or NuGet.

2. Sign in

Authenticate once with your Tempr license key — the same one you use in the IDE:

tempr auth

3. Add a key, pick a model

Add your own key for a provider, then set a default model; override it per-prompt whenever you like:

tempr keys add
tempr config set-model <id>

4. Put it to work

One-shot a task, or open the REPL:

tempr "explain this repo"
tempr

Command	Does
`tempr "<prompt>"`	One-shot agent turn with real tool-calling
`tempr`	Interactive REPL with persistent history and tab-completion
`<something> \| tempr`	Headless/piped mode — reads the prompt from stdin, prints to stdout
`tempr --ask "<prompt>"`	Plain chat, no tools
`tempr "@debugger <prompt>"`	Run a specialist persona
`tempr config`	Manage settings and your default model
`tempr history`	Browse saved sessions
`tempr mcp`	Add, list, and manage MCP servers
`tempr doctor`	Check connectivity and diagnose your setup

Run tempr --help for the full command reference — or read it in the CLI docs, including the scripts, pipes & CI guide.

Direct downloads

Every build is self-contained — no .NET runtime required. These links always serve the current release; each has a .sha256 beside it to check the download against.

Package managers

The same release, through a tool you already use. Update it the same way you installed it — tempr update tells you which.

Package manager	Install
npm Windows, macOS, Linux	`npm i -g @tempr/cli`
Homebrew macOS, Linux	`brew tap chrissalomon/tempr` `brew trust chrissalomon/tempr` `brew install tempr`
Scoop Windows	`scoop bucket add tempr https://github.com/chrissalomon/scoop-tempr` `scoop install tempr`
NuGet .NET global tool	`dotnet tool install -g Tempr.Cli` — needs the .NET SDK

Same keys, same models

The same ~30-provider allowlist as Tempr Chat — your key, your model, straight from the terminal. A few of the most popular:

OpenAIFirst-party

AnthropicFirst-party

Google GeminiFirst-party

DeepSeekFirst-party

xAIFirst-party

MistralFirst-party

Meta AIFirst-party

CohereFirst-party

OpenRouterAggregatorChat & CLI

Together AIInference

Fireworks AIInference

Hugging FaceInference

View all providers & models →

No separate subscription

The CLI runs on your existing Tempr Chat license — the same account, provider keys, and models you already use in the IDE. Every paid plan includes a 7-day free trial, and the CLI comes with all of them.

$15 / moStandard — CLI included

$20 / moPro — CLI included

From $35 / seatTeams — CLI included

See Chat pricing — CLI included

CLI FAQ

The questions specific to the terminal. Account, licensing, and provider questions live on the general FAQ page.

Nothing. The Windows, macOS, and Linux builds are self-contained, so curl -fsSL https://temprhq.io/cli/install.sh | sh (or irm https://temprhq.io/cli/install.ps1 | iex on Windows) is all it takes — there's no runtime to install first. You can also download a build directly and put tempr on your PATH.

Yes — it drives the same tool set as the IDE: file read, write, and edit, run_command, and semantic code search. It's a real agent loop, not a chat wrapper.

Yes. Piped mode reads the prompt from stdin and writes to stdout like any other command-line tool — { echo "review this:"; git diff; } | tempr — and --json, -y, and TEMPR_LICENSE_KEY cover unattended runs, so you can drop it into scripts, build steps, and scheduled jobs.

The same built-in lineup as the IDE — @debugger, @test, @review, @security, and the rest — invoked as tempr "@debugger why does this fail". Tab-completion in the REPL lists them all.

Also on your account

Same agent, same keys, same login — wherever you're working.

The agent in your IDE — Tempr Chat An OpenAI-compatible API — Tempr Gateway