Kilo Code

kilo.ai · Compute, Inference · 2 records · JSON

Kilo Code Computecaveatemailno cardD
Kilo Gateway Inferencecaveatanonymousno cardMrequest-rate

Kilo Code Compute

caveatemailno cardD

kilo.ai/pricing · cloud VM · compute/kilo-code · JSON · compare

Free tier	Kilo Code: "Platform plan \| Free for individuals" and for inference "Choose free models, local models, BYOK, Kilo Gateway, or Kilo Pass", with "AI inference stays at provider rates with no markup", quoted unit none; metered on wall-clock; free-product
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	email not published
Binding	none nothing binds on the platform itself; the ceiling is whichever model the free-model route serves that hour.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	a free VS Code agent with a free-model route is a $0 front end to whichever inference vendor is giving tokens away, so it inherits that vendor's quota and needs no card of its own.
Reliability	no reliability event recorded
Notes	The page does not name which models the free route serves or how many requests it allows. kilo.ai/docs/code-with-ai/platforms/free-models returned 404, so the number that would settle this record is not published at the URL I tried.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (2)

date	source	what it proved
2026-10-22	kilo.ai/pricing	"Platform plan \| Free for individuals" and "AI inference \| Choose free models, local models, BYOK, Kilo Gateway, or Kilo Pass", quoted
2026-10-22	kilo.ai/pricing	cloud compute is a separate metered line, quoted as "Usage-based compute for cloud features, charged separately from inference"

More fields (1)

{
 "wave2": {
  "class": "free-standing-agent",
  "free_tier_kind": "free-product",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": null,
   "gpu_count": null
  }
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

Kilo Gateway Inference

caveatanonymousno cardMrequest-rate

api.kilo.ai/api/gateway · inference gateway, keyless · inference/kilo-gateway · JSON · compare

Free tier	14 of the 401 ids in the catalogue end in ':free' and require no key; three more are free without the suffix (reflection/beam-free, kilo-auto/free, openrouter/free). Eight of the nine ids probed answered in both runs unit tokens; metered on request-count; the catalogue is 542343 bytes and was counted by id: 401 ids, 14 ending in ':free'. The prober's '19 of 401 answering' is its own figure and is quoted as third-party, not as a count made here
Price	none stated
Per-object cap	not applicable: an inference endpoint stores nothing for the caller
Retention	not applicable
Egress	the response is the payload; no separate egress term is published
Identity	anonymous no key for the :free ids. The prober's per-endpoint track record reports 549 of 1,464 calls answered over 10 days, 38 per cent, and 549 of the 685 calls Kilo accepted for a model it serves, 80 per cent
Limits	request-rate unpublished. One `:free` id returned 404 from this IP in both runs while the prober had it at 34 of 35, so the ceiling is per-IP and per-tier
Binding	request-rate the per-IP tier binds: an id that answers elsewhere 404s here, which is a ceiling on the caller rather than on the endpoint
Interface	rest endpoint: https://api.kilo.ai/api/gateway/chat/completions
Tooling	any OpenAI-compatible client; curl against POST {base}/chat/completions
Composes with	pairs with an agent as the widest free catalogue; the scarce link is the per-IP tier, because an id that answers elsewhere 404s from here
Reliability	one :free id that the prober had at 34 of 35 returned 404 from this IP in both runs, with the upstream provider's own error body inside the gateway's
Notes	openrouter/free is a router id whose served model changes between calls, which makes it the least predictable row in the sweep and the clearest illustration that an echoed model string must be recorded per call rather than per endpoint.
Evidence rung	measured eight ids were called twice. The catalogue size is third-party because the 540 KB listing was not enumerated.
Checked	2026-10-23 (live); stale on 2026-11-22

Evidence (4)

date	source	what it proved
2026-10-22	probe:https://api.kilo.ai/api/gateway/chat/completions	liquid/lfm-2.5-2.6b:free answered 200 with the model echoed, 98 completion tokens, 11702 ms in run 1 and 5177 ms in run 2; cohere/north-mini-code:free answered with 88 tokens at 5721 ms and 3235 ms
2026-10-22	probe:https://api.kilo.ai/api/gateway/chat/completions	openrouter/free echoed nvidia/nemotron-3.5-lightning:free in run 1 and nvidia/nemotron-3.5-content-safety in run 2, so the same id routed to two different models eight minutes apart
2026-10-22	probe:https://api.kilo.ai/api/gateway/chat/completions	nvidia/nemotron-3-super-120b-a12b:free returned 404 in both runs; the body nests the upstream provider's 404 rather than Kilo's
2026-10-22	api.kilo.ai/api/gateway/models	401 model ids in a 542343-byte payload at HTTP 200; 14 ids end in ':free'. The ':free' suffix is not the only free marker: reflection/beam-free, kilo-auto/free and openrouter/free carry no suffix

Unknown or withheld values (10)

$.probe.records[0].tps_basis process
$.probe.records[1].tps_basis process
$.probe.records[2].tps_basis process
$.probe.records[3].tps_basis process
$.probe.records[4].tps_basis process
$.probe.records[5].tps_basis process
$.probe.records[6].tps_basis process
$.probe.records[7].tps_basis process
$.probe.records[8].tps_basis process
$.retention.stated not-applicable