asale-sell-endpoint

Sell a custom OpenAI-compatible endpoint on the asale market: probe it, connect it as an account, and price it above what its own tokens cost. 把自定义 OpenAI 兼容端点挂到 asale 市场上卖：先探测、接入成账号，并把价格定在它自己的 token 成本之上。

English · 中文

Sell a custom OpenAI-compatible endpoint on the asale market: probe it, connect it as an account, and price it above what its own tokens cost.

Trigger Keywords

• sell an API endpoint
• resell an OpenAI-compatible API
• connect a custom endpoint to asale
• asale sell endpoint

Authentication

Two separate things have to be true.

1. The local daemon must be running, and you need its token. asaled writes ~/.asale/daemon.token (mode 0600) on first run and requires it on every /rpc call, loopback included.

bash

asale status            # is the daemon up, and on which port
asale start             # start it if it is not

bash

# Every /rpc call carries the daemon token — loopback included.
TOKEN=$(cat ~/.asale/daemon.token)
call() {
  body=$2; [ -n "$body" ] || body='{}'
  # --noproxy is not optional: a machine with HTTP_PROXY set sends even
  # 127.0.0.1 through it, and the proxy answers 502 with an empty body,
  # which reads exactly like a broken daemon.
  curl -sS --noproxy 127.0.0.1 -X POST "http://127.0.0.1:9700/rpc/$1" \
    -H 'content-type: application/json' -H "x-asale-token: $TOKEN" -d "$body"
}

A connection refused means the daemon is not running — say so and stop. Starting it is the user's call.

~/.asale/asaled.bind holds the port the last asale start used; fall back to 127.0.0.1:9700. A bind on 0.0.0.0 is not a destination — keep the port and dial loopback.

2. Selling and buying need a signed-in asale account. The daemon mints the consumer key and registers the seller against that session. There is no CLI login: open the app (asale open) and sign in there. Without it you get errors.session.signInToSell / errors.session.signInToBuy, which is not something to work around.

About & Provenance

• Source: github.com/asale-ai/asale
• Homepage: asale.ai
• Install: curl -fsSL https://asale.ai/dl/install.sh | sh (Windows: irm https://asale.ai/dl/install.ps1 | iex). asale update re-runs the same installer.
• Config: ~/.asale/ — the token, the SQLite store and the daemon's logs. $ASALE_DATA_DIR moves all of it.

This skill talks to a daemon on your own machine. Nothing here reaches the asale servers directly; the daemon does that, over its own authenticated session. The token file is the only credential this skill reads, and it never leaves the loopback interface.

How It Works

Selling is per subscription account, not per machine. The daemon holds its own copy of the account's credential (it never reads or writes the CLI's config on this side), declares the account's models to the market as lanes, and serves other people's requests from it — inside the limits you set: a price floor, how many requests at once, and an optional daily token cap.

A lane leaves the market on its own whenever the market price drops under your floor, and comes back when it recovers. Nothing has to be switched off and on again for that.

A custom endpoint is sold as an account like any other, except that its credential is an API key you supply rather than a subscription this machine was already signed into. It is also the one seller whose tokens have a marginal cost, which is what the price floor has to be set against.

Usage

1. Connect the endpoint

bash

call connect_custom_endpoint '{"baseUrl":"https://api.example.com/v1","apiKey":"<key>","label":"house","minRatio":10,"concurrency":5}'

The daemon validates the URL and probes GET {base}/models with the key before storing anything, so a wrong host or a dead key fails here rather than on the first buyer. wire is optional — left out, the probe tries each dialect and keeps the first that answers, which is the right default because the moment of connecting is when its operator is least sure ("OpenAI-compatible" is written on hosts that also serve /messages).

The reply names what happened:

Field	Means
account_id	The account id every later call names it by.
wire	The protocol the endpoint actually answered on.
endpoint_models	How many models it serves.
sellable_models	How many of those the market trades.

Re-running it with the same label updates that account in place — endpoint, key and terms are all rewritten and the cached model list is replaced.

2. Review and adjust the terms

bash

call list_custom_endpoints
call custom_endpoints_status
call set_account_sell '{"provider":"custom","accountId":"<id>","enabled":true,"minRatio":30,"concurrency":10}'

The price floor

minRatio is a whole percentage of the vendor's list price — 60 means "I will not sell below 60% of list". 100 is list price.

• 5 is the platform's floor. The market never prices anything below it, so a floor of 5 never withholds anything.
• 10 is the default a fresh account starts on.

There is no "any price" setting: every account trades against a floor. Lowering one is a real decision about money — propose it, do not just do it.

A metered endpoint is the one case where the floor is not only about profit: set it under what your own tokens cost and every sale loses money. Work the number out from the endpoint's own pricing before proposing it.

3. Refresh or remove

bash

call refresh_custom_endpoint '{"accountId":"<id>"}'   # re-probe the model list
call set_account_sell '{"provider":"custom","accountId":"<id>","enabled":false}'

CRITICAL INSTRUCTION FOR AGENT:

1. connect_custom_endpoint spends the user's key on a probe request and stores it. Confirm the URL and the key with the user before calling it — never as an incidental step of something else.
2. Never call remove_custom_endpoint. Switching selling off is reversible; removing the account is not.
3. Set minRatio from the endpoint's real cost per token, not from the default. The default exists for subscriptions, where the marginal token is already paid for.
4. Read sellable_models back to the user. "Connected, 400 models, selling 12" is the answer they need; "connected" alone is not.

Error Handling

Code	Error Type	Example Message
—	Daemon not running	curl: (7) Failed to connect to 127.0.0.1 port 9700
401	Bad or missing token	{"key":"errors.daemon.unauthorized","message":"unauthorized (missing or bad X-Asale-Token)"}
400	Not signed in (selling)	{"key":"errors.session.signInToSell","message":"sign in before selling"}
400	Not signed in (buying)	{"key":"errors.session.signInToBuy","message":"sign in before buying"}
400	Unknown account	{"message":"unknown account"}
400	Bad tool id	{"message":"unknown tool: <id>"}

AGENT CRITICAL INSTRUCTION:

1. Errors carry a key as well as a message. The key is a stable translation id — quote it, do not paraphrase the message.
2. On errors.session.signInToSell / errors.session.signInToBuy, tell the user to sign in from the app (asale open) and stop. Do not retry, and do not look for another route to the same effect.
3. On a connection failure, report that the daemon is down and let the user decide whether to start it. Do not run asale start unasked.

Tips

Visit https://asale.ai for more information. The same switches have a UI in the desktop app and in any browser — asale open opens it, and anything this skill does can be checked there.