Requesty CLI

Point Claude Code, Codex, OpenCode, Pi, Hermes and DeepSeek Harness at Requesty from one terminal app.

requesty finds the AI coding harnesses installed on your machine and rewrites their own configuration so every request goes through the Requesty gateway. You get 300+ models behind a single API key, one bill, and full logs and analytics for every tool you code with. It also shows what you are spending, right in the terminal. There is no proxy to keep running and no wrapper command to remember: your harnesses keep starting the way they always did.

Every file is backed up before it is touched, and existing settings are merged rather than replaced by default.

Contents:

Quick start
First run
Launching a harness
Profiles
Supported harnesses
Usage
Merge or overwrite
Backups and how to revert
Keys
Configuration
Advanced installation options
Upgrade
Uninstall
Download a release manually
Build from source
Troubleshooting

Quick start

curl -fsSL https://raw.githubusercontent.com/requestyai/cli/main/install.sh | sh

The installer adds ~/.requesty/bin to your PATH and tells you which shell file it changed. Open a new terminal, or run the source command it prints, then run requesty.

First run

Sign in with your browser

The first time you run requesty or a harness command such as requesty claude on a machine without a profile, the CLI opens your browser to sign in to Requesty straight away; once you approve, it creates an API key in your account, named after this machine (Requesty CLI (my-laptop)), and saves it as a profile in ~/.requesty/config.json. The app then carries on to its dashboard, or the harness you asked for starts. Every later run reuses that profile. The key shows up on the API keys page like any other, and can be revoked there.

╭──────────────────────────────────────────────────────────────────╮
│ Welcome to Requesty                                              │
│                                                                  │
│ One gateway for every model, in every tool you use, or app you   │
│ build.                                                           │
│                                                                  │
│ Waiting for you to finish signing in with your browser…          │
│                                                                  │
│ If it did not open, visit this address:                          │
│ https://app.requesty.ai/oauth/authorize?...                      │
│                                                                  │
│ Claude Code starts as soon as the key is saved.                  │
│                                                                  │
│ Working over SSH, or already have a key? Quit and run `requesty  │
│ login --api-key <key>` instead.                                  │
│                                                                  │
│ esc quit                                                         │
╰──────────────────────────────────────────────────────────────────╯

esc quits; run the command again to retry. The key is created in your group. If you belong to several groups a dialog asks which one; if you belong to none it is a personal key. Organizations that require keys to live in a group will say so, in which case ask an admin to add you to one. The profile is named default; pass --profile <name> to choose another name.

requesty login

requesty login runs the same sign-in from the command line. Use it to set up a machine before launching anything, to pick a group without being asked (--group Engineering), to name the profile yourself (--profile eu), to point it at another router (--router-url), or to replace a key that was revoked or has expired. Existing profiles are left alone unless you pass --force.

Over SSH, or with a key you already have

The browser hands the sign-in back to the CLI on 127.0.0.1, which does not work over SSH. Create a key on the API keys page and run requesty login --api-key <key> instead. The key is checked against the gateway before it is saved, so a typo is rejected here instead of failing later. It is saved as the profile default unless you pass --profile.

Launching a harness

requesty claude, requesty codex, requesty opencode, requesty pi and requesty hermes start the harness with its traffic routed through Requesty for that run only; the harness's own configuration is not changed. The current profile is used unless you select another.

The first time you launch a harness from a profile, the CLI settles which model it should use and remembers the answer in the profile. When the key can route to the harness's recommended default (claude-sonnet-5 for Claude Code, gpt-6-sol for Codex, and for OpenCode, Pi and Hermes whichever of the two the key can route to first), that is used and printed, without asking. Otherwise a picker asks: the Policies tab lists Requesty's managed policies, which name a model once and route it across providers, and tab switches to every model the key can route to. --model overrides the remembered model for one run without changing it; --choose-model opens the picker to change it. In a script or CI with no terminal to ask in, the default is still used when it can be; otherwise pass --model, or pick once interactively first.

Claude Code hands background work (session titles, summaries, subagents marked haiku) to a smaller model, which it would otherwise ask for by an Anthropic model id the gateway does not know. The CLI settles this the same way: claude-haiku-4-5 when the key can route to it, else the main model itself, so an access list without Haiku still works. --choose-fast-model opens the picker to change it and --fast-model overrides it for one run.

requesty claude                                    # current profile, remembered model (settled the first time)
requesty claude --profile eu                       # a named profile, with its own remembered model
requesty claude --model anthropic/claude-opus-4-1  # this model, just for this run
requesty claude --choose-model                     # pick again and remember the new answer
requesty claude --choose-fast-model                # pick what background work runs on
requesty codex --full-auto                         # everything else goes to the harness untouched
requesty claude --effort high                      # so its own flags work as usual
requesty opencode --model gpt-6-sol run "explain this repo"
requesty pi --thinking off
requesty hermes chat -q "hello"

--profile <name> picks the profile for a run; without it, the current profile set with requesty profiles use is used. For a harness command, put --profile after the harness name and before the first argument that is passed through to the harness.

Each harness is pointed at Requesty in the way it allows without touching its files:

Claude Code gets ANTHROPIC_BASE_URL and the key through the environment plus a --settings document for this run.
Codex gets a requesty model provider through -c overrides.
OpenCode knows Requesty as a built-in provider that switches on when REQUESTY_API_KEY is set; an OPENCODE_CONFIG_CONTENT document, merged into any you already export, pins it to this profile, and the model is passed as -m requesty/<model>. A Requesty key stored earlier with opencode auth login can take precedence over the profile's; the CLI warns when the two differ.
Pi has no flag for a custom provider, so the CLI writes a small extension and a catalog of the models your key can route to under ~/.requesty/pi/ and starts Pi with --extension. The catalog is refreshed from your account on each launch and kept from last time when that fails. Other vendors' API key variables are dropped from Pi's environment so every model it offers goes through Requesty.
Hermes is started as --provider custom with CUSTOM_BASE_URL pointing at the router. Hermes loads ~/.hermes/.env over the environment, so the CLI warns when a value there would replace one of its own.

Anything the harness itself can be told, such as reasoning effort, is left to the harness's own flags: claude --effort high, codex -c model_reasoning_effort=high, opencode --variant high, pi --thinking high, hermes --reasoning high. A model or provider flag you pass through takes precedence over the CLI's.

Profiles

A profile is one sign-in: an API key and the router it is sent to. Most people have one. You want more when your organization gives you keys in different groups with different model policies (an EU group and a US group, say), or when you use Requesty for work and personally on the same machine. Keys with manage permissions for the api-keys, groups and access-lists subcommands fit here too: save one with requesty login --api-key <key> --profile manage and pass --profile manage to those commands.

requesty profiles list                                  # what is saved, current marked with *
requesty profiles use eu                                # make this profile current
requesty profiles remove personal                       # forget a profile (the key is not revoked)

requesty login creates the default profile unless you pass --profile. The first profile becomes current. Every command accepts --profile <name>.

Harnesses configured from the dashboard stay pinned to the profile used during configuration. Claude Code stores that profile's key in its settings. Codex fetches the key through requesty auth token --profile <name>. Configure the harness again to move it to another profile.

Pick a harness

Harnesses found on your PATH are listed first. The ones that are not installed stay dimmed and cannot be configured. The footer always shows the keys available on the current screen.

  harnesses on this machine                    0 of 6 routing through requesty

        HARNESS             CONFIG                                  STATUS
  ────────────────────────────────────────────────────────────────────────────
  ❯ [ ] Claude Code         /home/you/.claude/settings.json         inactive
    [ ] Codex               /home/you/.codex/config.toml            inactive
    [ ] OpenCode            /home/you/.config/opencode/opencode.jsoninactive
    [ ] Pi                  /home/you/.pi/agent/models.json         inactive
    [ ] Hermes              /home/you/.hermes/config.yaml           inactive
    [ ] DeepSeek Harness    /home/you/.dsh/settings.yaml            inactive

  ╭──────────────────────────────────────────────────────────────────────────╮
  │ Claude Code                              /home/you/.claude/settings.json │
  │ takes a backup of settings.json                                          │
  │ writes a settings.json to route through Requesty                         │
  ╰──────────────────────────────────────────────────────────────────────────╯

  ↑/↓ move · space configure · r refresh · q/esc quit

Choose a model, then choose how to write the config

The model list is fetched from your account with your key, so it reflects the models and policies you actually have access to. Then choose merge (recommended) or overwrite. The CLI writes the files, the row flips to [✓] active, and the header count goes up.

Restart the harness

Harnesses read their configuration at startup. Restart claude, codex, opencode, pi, hermes or dsh and it is talking to Requesty.

Supported harnesses

Harness	Detected by	Files written	What Requesty sets
Claude Code	`claude` on `PATH`	`~/.claude/settings.json`	`ANTHROPIC_BASE_URL`, `ANTHROPIC_AUTH_TOKEN` and `ANTHROPIC_MODEL` in the `env` block
Codex	`codex` on `PATH`	`~/.codex/config.toml`	A `requesty` model provider on `.../v1`, the selected model, and command-backed auth through `requesty auth token`
OpenCode	`opencode` on `PATH`	`~/.config/opencode/opencode.json`	A `requesty` provider on `.../v1` plus the model as `requesty/<model>`
Pi	`pi` on `PATH`	`~/.pi/agent/models.json`	A `requesty` provider using the native Anthropic Messages API
Hermes	`hermes` on `PATH`	`~/.hermes/config.yaml`	A `requesty` entry in `custom_providers` and `model.default`
DeepSeek Harness	`dsh` on `PATH`	`$DSH_HOME/settings.yaml`, `$DSH_HOME/.credentials.yaml` (`~/.dsh` by default)	A `requesty` provider on `.../v1` under `llm-pi-ai`, the selected model as `agent-default-model`, and the key stored as `REQUESTY_API_KEY`

Pi and Hermes are configured against the native Anthropic Messages format, which is what lets Requesty apply automatic prompt caching to those harnesses.

Where a harness supports custom headers, the CLI also sets an X-Title header naming the tool, so the Requesty dashboard can break spend down per harness.

DeepSeek Harness needs the models of a custom provider listed in its own settings, so the CLI writes the model you selected. You can add more Requesty models later inside the harness, under Settings, Models, Requesty, either with Fetch available models or with Add model. In merge mode a later run of the CLI keeps those entries and only adds the model you selected if it is missing.

Usage

Above the harness list, the CLI shows spend, requests and tokens for the last 30 days for the profile it is running as, and refreshes them on demand. Full breakdowns by model, user and tool live in the Requesty dashboard.

Merge or overwrite

The last step of the wizard asks how to write the configuration.

Merge existing config files (recommended). Your settings are kept and only the Requesty values are added or updated. Comments and formatting of TOML and YAML files are not preserved, because the files are re-encoded.
Overwrite config files. The file is replaced with only what Requesty needs. Use this when a config has drifted and you want a clean, known-good state.

Either way, a backup is taken first.

Backups and how to revert

Before the first write, each file is copied to <file>.requesty.bak, keeping its original permissions. The backup is only created once per file, so your true pre-Requesty state is never overwritten by a later run.

To go back to your original setup:

mv ~/.claude/settings.json.requesty.bak ~/.claude/settings.json

Keys

Your Requesty API keys are written into ~/.requesty/config.json, one per profile. Codex retrieves the current profile's key when needed through requesty auth token; other harnesses may also store it in their own config because that is how they authenticate. All files containing a key are written so that only your user can read them.

A key created by signing in can make completions and read its own usage, nothing more. The management subcommands (requesty api-keys, requesty groups, requesty access-lists) need a key with manage permissions from the API keys page, saved as its own profile with requesty login --api-key <key> --profile <name>. The short-lived token from the browser sign-in is used once to create the key and is never stored.

Treat those files as secrets and do not commit them. Keys can be rotated or revoked at any time on the API keys page; run requesty login afterwards to get a new one. Removing a profile with requesty profiles remove forgets the key locally only.

Configuration

~/.requesty/config.json:

{
	"current": "engineering",
	"profiles": {
		"engineering": {
			"api_key": "rqsty-..."
		},
		"eu": {
			"api_key": "rqsty-...",
			"router_base_url": "https://router.eu.requesty.ai"
		}
	}
}

current names the profile used when none is selected explicitly. It is required whenever any profiles are saved.
profiles.<name>.api_key authenticates API and harness requests.
profiles.<name>.router_base_url is that profile's inference endpoint. When omitted, it defaults to https://router.requesty.ai.

Set a profile's router_base_url to use a different region or self-hosted deployment. The config has no legacy single-key format; delete the file to start over from onboarding.

Advanced installation options

The installer downloads the latest release, verifies its checksum, installs the binary into ~/.requesty/bin, and adds that directory to your PATH in .zshrc, .bashrc or config.fish. Open a new shell, or source your shell config, and requesty is on your PATH.

It accepts a few options, for example to pin a version or install somewhere else:

curl -fsSL https://raw.githubusercontent.com/requestyai/cli/main/install.sh | sh -s -- --version v1.2.3
curl -fsSL https://raw.githubusercontent.com/requestyai/cli/main/install.sh | sh -s -- --install-dir ~/.local/bin
curl -fsSL https://raw.githubusercontent.com/requestyai/cli/main/install.sh | sh -s -- --no-modify-path

The same settings are available as environment variables: REQUESTY_VERSION, REQUESTY_INSTALL_DIR and REQUESTY_HOME. Run install.sh --help for the full list.

Upgrade

Run the install command again. The binary is replaced in place and everything in ~/.requesty, including config.json, is preserved:

curl -fsSL https://raw.githubusercontent.com/requestyai/cli/main/install.sh | sh

Uninstall

Restore any harness config you want back from its .requesty.bak backup, then:

rm -rf ~/.requesty

Remove the block marked # >>> requesty cli installer >>> from your shell configuration file to take the install directory back off your PATH.

Download a release manually

Platform	Command
macOS (Apple Silicon)	`curl -fLO https://github.com/requestyai/cli/releases/latest/download/requesty_darwin_arm64.tar.gz`
macOS (Intel)	`curl -fLO https://github.com/requestyai/cli/releases/latest/download/requesty_darwin_amd64.tar.gz`
Linux (ARM64)	`curl -fLO https://github.com/requestyai/cli/releases/latest/download/requesty_linux_arm64.tar.gz`
Linux (x86_64)	`curl -fLO https://github.com/requestyai/cli/releases/latest/download/requesty_linux_amd64.tar.gz`
Windows (ARM64)	`curl.exe -fLO https://github.com/requestyai/cli/releases/latest/download/requesty_windows_arm64.zip`
Windows (x86_64)	`curl.exe -fLO https://github.com/requestyai/cli/releases/latest/download/requesty_windows_amd64.zip`

Unpack the archive and put the requesty binary anywhere on your PATH.

Build from source

The Go version in go.mod is what CI builds with.

go run .            # run the CLI
go build -o requesty .
go test ./...
gofmt -l .          # CI fails if this prints anything

Troubleshooting

A harness is listed but cannot be configured. Its executable was not found on your PATH. Install the harness, then refresh the list.

that API key was not recognised. The gateway does not accept the key you gave requesty login --api-key. Copy it again from the API keys page.

The browser never comes back to the terminal. The sign-in finishes by redirecting your browser to 127.0.0.1 on the machine running the CLI, so it cannot complete over SSH or in a container. Run requesty login --api-key <key> with a key from the API keys page instead, or forward the port shown in the sign-in URL.

no Requesty profile configured; run requesty login. A command that cannot onboard automatically ran without a saved profile. Run requesty login or requesty login --api-key <key>.

group_id is required for organizations in group budget mode. Your organization tracks spend per group, and you are not in one yet. Ask an organization admin to add you to a group, then sign in again.

A harness gets a 401. The key was revoked or has expired. Replace that profile with requesty login --profile <name> --force. With OpenCode, also check for a Requesty key stored by opencode auth login, which wins over the profile's; opencode auth logout removes it. With Hermes, check ~/.hermes/.env for a REQUESTY_API_KEY or CUSTOM_BASE_URL entry, which Hermes loads over the environment.

Requesty needs pi 0.84.0 or newer. requesty pi registers Requesty through Pi's provider extension API, which arrived in Pi 0.84.0. Upgrade Pi and try again.

Could not load usage. The usage panel asks the management API about your own key, which any key may do. A 401 means the key was revoked or has expired: run requesty login to get a new one. Routing itself is unaffected by this panel.

The harness still uses its old provider. Restart it. Configuration is read at startup.

Name		Name	Last commit message	Last commit date
.github		.github
cmd		cmd
internal		internal
.gitignore		.gitignore
.goreleaser.yaml		.goreleaser.yaml
LICENSE		LICENSE
README.md		README.md
go.mod		go.mod
go.sum		go.sum
install.sh		install.sh
main.go		main.go

Navigation Menu

Repository navigation

About

Repository files navigation

Requesty CLI

Quick start

First run

Launching a harness

Profiles

Supported harnesses

Usage

Merge or overwrite

Backups and how to revert

Keys

Configuration

Advanced installation options

Upgrade

Uninstall

Download a release manually

Build from source

Troubleshooting

Learn more

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

Requesty CLI

Quick start

First run

Launching a harness

Profiles

Supported harnesses

Usage

Merge or overwrite

Backups and how to revert

Keys

Configuration

Advanced installation options

Upgrade

Uninstall

Download a release manually

Build from source

Troubleshooting

Learn more

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages