{% extends "base.html" %} {% block title %}Accounts · modelpass bench{% endblock %} {% block body %} {% if note %}
{{ note }}
{% endif %}

Configured accounts

New account

From {{ store_path }}. Each profile combines a vendor account with its storage boundary. Clients select the stable ID; people see the nickname.

{% if group_filter %}
Showing the accounts in group {{ group_filter }}. Show all accounts · Groups
{% endif %} {% for n in store_notes %}
{{ n }}
{% endfor %} {% if not connections %}

No accounts configured.

A fresh install has no AI connectivity at all — that is the design, not a missing step. Add one with the button above, or on the command line with modelpass connect anthropic.

{% endif %} {% for c in connections %}

{{ c.display_name }} {% if c.nickname %}id: {{ c.name }}{% endif %} {% if not c.enabled %}disabled{% endif %} {% if c.auth_mode == 'subscription' %} subscription {% if c.identity_verified %} identity verified {% else %} identity not verified {% endif %} {% else %} metered api_key {% endif %} {% if not c.guards_configured %}no spend guards{% endif %} {% for g in c.groups %}{{ g }}{% endfor %}

Check Edit
{% if c.auth_mode == 'subscription' %}
{% elif c.is_compatible %} {# The one button on this page that spends, so it says so before it does: three short calls against whatever the base URL points at. #}
{% endif %}
vendor
{{ c.vendor }}
stable ID
{{ c.name }}
runtime
{{ c.runtime }}
auth mode
{{ c.auth_mode }}
groups
{{ c.groups | join(', ') }} {% if not c.declared_groups %} — this account names no group, which is what puts it in default. {% endif %}
credentialRef
{{ c.credential_ref }} — {{ c.credential_describes }} {% if not c.credential_present %} not set in this environment {% endif %}
{% if c.base_url %}
base URL
{{ c.base_url }}
{% endif %} {% if c.model %}
model
{{ c.model }}
{% endif %}
timeout
{% if c.timeout_seconds %} {{ c.timeout_seconds }}s — the default bound for calls here; a call that names its own replaces it rather than being clamped by it. {% else %} unbounded — no timeoutSeconds is set, so a call runs until it finishes or the caller passes a bound. {% endif %}
input window
{% if c.max_input_tokens is not none %} {{ "{:,}".format(c.max_input_tokens) }} tokens — what this account was told the model's usable input window is. modelpass does not check a payload against it; it is here for the tools that do. {% else %} unknown — no maxInputTokens is set. modelpass has no table of model window sizes and will not guess one, so nothing here can tell you whether a payload fits. {% endif %}
prompt cache
{% if c.prompt_cache %} requested: {{ c.prompt_cache }} — {{ c.prompt_cache_note }} {% else %} nothing stated — no promptCache is set. That is not “caching off”: most runtimes here cache prompts whether or not anyone asks, and this connection has simply not said anything about it. {% endif %}
retry
{% if c.retry == 'never' %} never — every retryable verdict on this connection reads no. modelpass never retries for you either way; this is what your own loop is told. {% else %} default — verdicts are classified from typed facts. modelpass never retries for you; the loop is yours. {% endif %}
{% if c.verified_any %}
verified capabilities
{{ c.verified.supported | join(', ') or 'nothing' }} {% if c.verified.unsupported %}
not supported: {{ c.verified.unsupported | join(', ') }}
{% endif %}
checked {{ c.verified.checked_at or 'at an unrecorded time' }} {% if c.verified.models %} — models seen: {{ c.verified.models | join(', ') }} {% endif %}
{% elif c.is_compatible %}
verified capabilities
Nothing yet. This runtime names an API shape, so every cell reads unverified — including chat — until a drive answers for your endpoint. Use Verify above.
{% endif %} {% if c.config_dir %}
config root
{{ c.config_dir }}
{% endif %} {% if c.setup_note %}
one-time setup
{{ c.setup_note }}
{% endif %} {% if c.verified_identity %}
pinned identity
{% if c.verified_identity.email %}{{ c.verified_identity.email }}{% endif %} {% if c.verified_identity.organization_id %} {% if c.verified_identity.email %}; {% endif %}org {{ c.verified_identity.organization_id }} {% endif %}
{% elif c.auth_mode == 'subscription' %}
identity protection
Not pinned. Verify this account before relying on its nickname.
{% endif %} {% if c.auth_mode == 'subscription' %}
live profile
This page reads only what is configured — it asks no vendor anything. Check runs the token-free identity probe and shows the account the runtime is logged into right now.
{% endif %}
guards
{% if c.guards_configured %} {{ c.guard_thresholds | join('; ') }} {% else %} no spend guards configured — nothing bounds what one run on this connection may spend. modelpass ships no default threshold because it has no basis for a number; see docs/guards.md. {% endif %}
on quota exhausted
{% if c.quota_failover %} fail over to {{ c.quota_failover }} {% else %} stop cleanly {% endif %}
scrubbed at launch
{% if c.scrubbed %}{{ c.scrubbed | join(', ') }} {% else %}nothing on the scrub list is set right now{% endif %}
{% if c.kept %}
kept
{{ c.kept | join(', ') }}
{% endif %} {% if c.allow_env %}
allowEnv
{{ c.allow_env | join(', ') }}
{% endif %} {% if c.description %}
note
{{ c.description }}
{% endif %} {% if c.plan_problem %}
problem
{{ c.plan_problem }}
{% endif %}
{% endfor %} {% if receipt %}

Preflight receipt — {{ receipt.connection }}

{% if receipt.failed %}
The preflight could not run.
{{ receipt.failed }}
{% else %}

{{ receipt.summary }}

runtime
{{ receipt.runtime }}
requested mode
{{ receipt.requested_auth_mode }}
detected mode
{% if receipt.detected_auth_mode %} {{ receipt.detected_auth_mode }} {% else %} declared, not confirmed — the runtime did not report a mode {% endif %}
credential source
{{ receipt.credential_source or '—' }}
{% if receipt.account %}
account
{{ receipt.account }}
{% endif %} {% if receipt.plan_name %}
plan
{{ receipt.plan_name }}
{% endif %}
identity
{% if receipt.identity_verified == true %}VERIFIED {% elif receipt.identity_verified == false %}MISMATCH {% else %}NOT PINNED{% endif %}
{% if receipt.account_profile %} {% if receipt.account_profile.organization_name %}
organization
{{ receipt.account_profile.organization_name }}
{% endif %} {% if receipt.account_profile.organization_id %}
organization ID
{{ receipt.account_profile.organization_id }}
{% endif %} {% if receipt.account_profile.auth_method %}
login method
{{ receipt.account_profile.auth_method }}
{% endif %} {% if receipt.account_profile.api_provider %}
API provider
{{ receipt.account_profile.api_provider }}
{% endif %} {% if receipt.account_profile.subscription_type %}
subscription type
{{ receipt.account_profile.subscription_type }}
{% endif %}
profile source
{{ receipt.account_profile.source }}
{% endif %}
scrubbed
{% if receipt.scrubbed %}{{ receipt.scrubbed | join(', ') }} {% else %}nothing to scrub{% endif %}
{% if receipt.preserved %}
kept (credential)
{{ receipt.preserved | join(', ') }}
{% endif %} {% if receipt.passthrough %}
kept (allowEnv)
{{ receipt.passthrough | join(', ') }}
{% endif %} {% for d in receipt.directives %}
directive
{{ d.name }} = {{ d.value }} — {{ d.reason }}
{% endfor %} {% for n in receipt.notes %}
note
{{ n }}
{% endfor %}
guards
{{ receipt.guard_note or 'configured' }}
{% if receipt.cache_note %}
cache
{{ receipt.cache_note }}
{% endif %}
status
{% if receipt.ok %}OK {% else %}FAILED — {{ receipt.problem }}{% endif %} {% if not receipt.enabled %} disabled — the preflight still runs; what enabled = false switches off is spending, not looking. {% endif %}
{% endif %} {% endif %}

Stored keys

{% if not secrets.exists %}

No secrets file at {{ secrets.path }}. Nothing is stored, which is the normal state for an install whose connections use a subscription login or a named environment variable.

{% else %}

From {{ secrets.path }}. Entry names and references only — no value is read into this page, and there is no button here that deletes a key: an orphaned entry is recoverable and a deleted one is not.

{% if secrets.permissions %}

{{ secrets.permissions }}

{% endif %} {% for n in secrets.notes %}
{{ n }}
{% endfor %} {% if not secrets.entries %}

The file exists and holds no entries.

{% else %}
{% for e in secrets.entries %}
{{ e.entry }}
{% if e.used_by %} used by {{ e.used_by | join(', ') }} {% else %} orphan no connection references this, so it is inert. It is listed, never collected for you. {% endif %}
{% endfor %}
{% endif %} {% endif %}

Ambient environment

These are the variables each runtime would scavenge if modelpass let it. Nothing here is a connection, and nothing here is used unless a connection names it — but a variable that is set is a variable a vendor runtime would have picked up on its own, so it is worth knowing about.

Names only. modelpass can see these without ever passing them on, which is the whole point — no value is read into this page.

{% for row in audit %}
{{ row.runtime }}
{% if row.present %} {% for name in row.present %} {{ name }}{% if not loop.last %} {% endif %} {% endfor %}
{{ row.sentence }}
{% else %} clean nothing on this runtime's scrub list is set here. {% endif %}
watched: {{ row.watched | join(', ') }}
{% endfor %}

This audit reflects the environment of the process serving this page. A terminal or editor started before you cleaned up your environment still carries its own copy of the old variables until it is restarted, so a freshly launched process can read "clean" here while a long-lived one three windows over does not. That divergence is exactly why modelpass scrubs at launch rather than trusting whatever environment it happens to inherit — the guarantee has to hold in the stale process too.

Capabilities

What each runtime can actually do, from the capability registry. unverified is not supported: it means nobody checked, and modelpass treats it as unusable rather than guessing.

{% for r in runtimes %}{% endfor %} {% for row in capabilities %} {% for cell in row.cells %} {% endfor %} {% endfor %}
capability{{ r }}
{{ row.capability }} {{ cell.support }} {% if cell.note %} {# Collapsed by default: the reasons are long because they are sourced, and a table nobody can scan is not honest either. #}
why
{{ cell.note }}
{% endif %}

Run log

{% if run_log_enabled %} Every completed run appends one line to {{ run_log_path }} — when, which connection, which auth mode actually paid for it, tokens, outcome. See the log. {% else %} The run log is switched off for this home ([settings] runLog = false). Completed runs are not being recorded, so "which of these was billed to a key" has no durable answer here. {% endif %}
{% endblock %}