Skip to content
Select themeSelect language

post_api_models_catalog_refresh

POST
/api/models/catalog/refresh
curl --request POST \
--url https://example.com/api/models/catalog/refresh
Media typeapplication/json
object
fetched_at
required
string | null format: date-time
last_error
string | null
providers
required
Array<object>
object
id
required
string
models
required
Array<object>
object
available_modes

Axis 2 (mig 272) — the model’s harness-native available_modes (Claude ['build','plan']), joined from model_capabilities. None until a probe has landed; the FE then hides the Mode picker (degrade).

Array<string> | null
context
integer | null format: int32
cost
One of:
null
id
required
string
modalities_input
required
Array<string>
modalities_output
required
Array<string>
name
required
string
reasoning
required
boolean
subagent_cap_level

Mig 358 — enforced | advisory | unsupported: whether a concurrency ceiling structurally holds on this harness (codex takes a real per-session cap) or is only stated to the model. The max-subagents control labels itself from this, so a wish is never displayed as a guarantee.

string | null
subagent_cap_note

Mig 358 — the measured detail behind [Self::subagent_cap_level].

string | null
subagent_evidence

Mig 358 — measured | documented: whether the claim was verified against the running CLI or read off the harness’s own docs.

string | null
subagent_reason

Mig 358 — WHY [Self::supports_subagents] is what it is, quoted verbatim from the runner’s harness registry. The FE renders this instead of inventing an explanation, so a hidden or disabled control always says why.

string | null
subagent_wired_by

Mig 358 — the seam the SUBAGENTS axis reaches this harness through. None while supports_subagents is Some(true) is the honest “the harness can, we do not drive it yet” state: the FE shows the toggle as a no-op rather than pretending it works or hiding a real capability.

string | null
supported_effort_levels

#626 D4.0 — the model’s HARNESS-NATIVE effort ladder (advertised order), joined from the model_capabilities table (the runner’s task-START capability probe). None until a probe has landed for this (provider, model_id) — the FE then degrades the EffortSlider to the canonical 5-step scale. The top stage (ultracode/ultra) is the last element when the harness presents it.

Array<string> | null
supports_fast_mode

#626 D4.0 — the model’s supports_fast_mode capability flag, joined from model_capabilities. None until a probe has landed.

boolean | null
supports_subagents

Axis 1 (mig 272; TRI-STATE since mig 358) — the model’s supports_subagents capability flag, joined from model_capabilities.

Some(true) — a probe established that the harness has subagents. Some(false) — a probe established that it genuinely has none; the FE disables the toggle and shows subagent_reason. None — no probe has landed. UNKNOWN: the FE offers the toggle with an “unverified” hint rather than hiding it, because hiding an unmeasured axis is indistinguishable from hiding an unsupported one — the ambiguity this tri-state removes.

(The former note “Some(true) only on Claude in v1” was already false when it was written and is gone: codex, opencode, openclaude, gemini/agy, qwen, qwen-code, openrouter and copilot all report true — eight harnesses, several of them measured against the pinned CLI. See harness-kit/subagents.ts.)

boolean | null
tool_call
required
boolean
name
required
string
refresh_interval_hours
required
integer format: int32
source
required
string
Examplegenerated
{
"fetched_at": "2026-04-15T12:00:00Z",
"last_error": "example",
"providers": [
{
"id": "example",
"models": [
{
"available_modes": [
"example"
],
"context": 1,
"cost": {
"cache_read": 1,
"cache_write": 1,
"input": 1,
"note": "example",
"output": 1
},
"id": "example",
"modalities_input": [
"example"
],
"modalities_output": [
"example"
],
"name": "example",
"reasoning": true,
"subagent_cap_level": "example",
"subagent_cap_note": "example",
"subagent_evidence": "example",
"subagent_reason": "example",
"subagent_wired_by": "example",
"supported_effort_levels": [
"example"
],
"supports_fast_mode": true,
"supports_subagents": true,
"tool_call": true
}
],
"name": "example"
}
],
"refresh_interval_hours": 1,
"source": "example"
}

Structured client error

Media typeapplication/json

The canonical JSON body of every error response — the single source of truth the frontend binds to. Every AppError serializes as this exact shape, and the generated OpenAPI component ApiErrorBody (with its ErrorCode enum) is what the frontend error schema is generated from, so there is no hand-written error schema on either end.

object
code
required

Machine-readable, stable error code.

string
Allowed values: not_found unauthorized forbidden license_required license_expired bad_request unprocessable precondition_failed conflict method_not_allowed rate_limited too_many_requests quota_exceeded database_error docker_error vault_error internal_error
details
One of:
null
error
required

Human-readable message (the server’s English text; the client may localize by code).

string
Example
{
"code": "not_found"
}

Authentication required

Media typeapplication/json

The canonical JSON body of every error response — the single source of truth the frontend binds to. Every AppError serializes as this exact shape, and the generated OpenAPI component ApiErrorBody (with its ErrorCode enum) is what the frontend error schema is generated from, so there is no hand-written error schema on either end.

object
code
required

Machine-readable, stable error code.

string
Allowed values: not_found unauthorized forbidden license_required license_expired bad_request unprocessable precondition_failed conflict method_not_allowed rate_limited too_many_requests quota_exceeded database_error docker_error vault_error internal_error
details
One of:
null
error
required

Human-readable message (the server’s English text; the client may localize by code).

string
Example
{
"code": "not_found"
}

Permission denied

Media typeapplication/json

The canonical JSON body of every error response — the single source of truth the frontend binds to. Every AppError serializes as this exact shape, and the generated OpenAPI component ApiErrorBody (with its ErrorCode enum) is what the frontend error schema is generated from, so there is no hand-written error schema on either end.

object
code
required

Machine-readable, stable error code.

string
Allowed values: not_found unauthorized forbidden license_required license_expired bad_request unprocessable precondition_failed conflict method_not_allowed rate_limited too_many_requests quota_exceeded database_error docker_error vault_error internal_error
details
One of:
null
error
required

Human-readable message (the server’s English text; the client may localize by code).

string
Example
{
"code": "not_found"
}

Structured server error

Media typeapplication/json

The canonical JSON body of every error response — the single source of truth the frontend binds to. Every AppError serializes as this exact shape, and the generated OpenAPI component ApiErrorBody (with its ErrorCode enum) is what the frontend error schema is generated from, so there is no hand-written error schema on either end.

object
code
required

Machine-readable, stable error code.

string
Allowed values: not_found unauthorized forbidden license_required license_expired bad_request unprocessable precondition_failed conflict method_not_allowed rate_limited too_many_requests quota_exceeded database_error docker_error vault_error internal_error
details
One of:
null
error
required

Human-readable message (the server’s English text; the client may localize by code).

string
Example
{
"code": "not_found"
}