Running coding agents unattended on a subscription: what each vendor actually says (Nov 2026)

Shared by Pavel · Updated 2026-11-11

Questions this answers

What is the documented way to run Claude Code unattended in CI on a subscription? Answerable in part (the route, not the rate or CLAUDE_CODE_ENTRYPOINT). https://github.com/anthropics/claude-code/issues/107106
Is `agy -p` metered differently from interactive Antigravity CLI use on Google AI Ultra? Answered by a vendor-repo collaborator. https://github.com/google-antigravity/antigravity-cli/issues/1411

Read date for everything below: 2026-11-11. Labels: [vendor] = vendor's own page or a collaborator on the vendor's repo; [measured: who, link]; [opinion: who]. "The vendor does not say" = no statement found on the pages read. "Not verified" = page could not be loaded.

1. Open questions on vendor repos, with URLs

#	Thread	Date	Exact question	Answered by vendor with a source?
1	anthropics/claude-code#107106 https://github.com/anthropics/claude-code/issues/107106	2026-11-04	"Is there a different rate for headless/SDK use, and if so, is it documented anywhere?" and "Is it permitted for an individual subscriber to set CLAUDE_CODE_ENTRYPOINT ... when running Claude Code headlessly for their own work?" and "is there an approved way to run unattended/automated sessions on a subscription at the same rate as the interactive CLI". 6 comments, all by the author (association NONE).	No. No Anthropic reply.
2	#107134 https://github.com/anthropics/claude-code/issues/107134	2026-11-04	"Is -p (--print) mode permitted for subscribers on the same basis as other official tools?" and "Does -p mode meter usage at an API-pricing-based / ~1.8x rate instead of the standard subscription rate?" plus a refund request. 5 comments, all the author.	No. No Anthropic reply.
3	#108211 https://github.com/anthropics/claude-code/issues/108211	2026-11-07	Bug/feature: headless `claude -p` silently waits through a usage limit and auto-resumes, so a supervising agent thought it had stopped; resumed runs duplicated work and burned ~50% of the weekly limit. Asks for a status line, an `--exit-on-limit` flag, and visible/cancellable paused runs. Reporter says support refused any compensation.	No. Two community replies (2026-11-08): fortitude-omnis (check `pgrep -fl "claude -p"`, pid file) and vijaym2k6 (wrap with `timeout 3600`; discloses he builds SteerPlane). [opinion: both, NONE association]
4	#106850 https://github.com/anthropics/claude-code/issues/106850	2026-11-03	Cloud routines run `claude -p --model haiku` subprocesses (no ANTHROPIC_API_KEY); ~140M tokens drained plan usage while the included Cloud session credit showed $248 of $250 left. Asks: document which meter a routine's usage bills to. 0 comments.	No.
5	#108851 https://github.com/anthropics/claude-code/issues/108851	2026-11-10	Feature request: a non-interactive way (-p / SDK / control protocol) to list and redeem usage-limit resets, as `/limit-reset` does interactively. 0 comments.	No.
6	#106161 https://github.com/anthropics/claude-code/issues/106161	2026-10-30	"Docs unclear: are background requests from a Claude Code mod allowed on a Pro/Max plan?" Asks (a) permitted for Free/Pro/Max, (b) any limit on frequency/duration without a new user message, (c) whether mods/hooks count as "explicitly permitted" under Consumer Terms. 0 comments.	No.
7	openai/codex discussion 59674 https://github.com/openai/codex/discussions/59674	2026-11-05	"Could the Codex team explicitly document ... whether interactive codex TUI, non-interactive codex exec, and codex app-server consume subscription quota at identical 1:1 rates, and commit to advance notice before introducing any mode-specific quota multipliers?" 1 comment (a follow-up by the same author on CLIProxyAPI and OpenAI Terms of Use).	No. No OpenAI reply.
8	google-gemini/gemini-cli discussion 30030 https://github.com/google-gemini/gemini-cli/discussions/30030	2026-10-29	"Is the official Gemini CLI allowed in a private automation tool on a Google AI Pro subscription?" (own server, own code, `gemini -p`, one account, no rotation; "Does that cover a personal script calling the official CLI? Or is the Gemini API the only supported route for this?") 0 comments.	No.

Additional threads found (last 14 days, gh search; search API hit a rate limit partway, so the sweep of openai/codex, gemini-cli and github/copilot-cli is incomplete = not verified for those repos):

#	Thread	Date	Question	Vendor answer?
A	google-antigravity/antigravity-cli#1411 https://github.com/google-antigravity/antigravity-cli/issues/1411	2026-11-04	Is `agy -p` metered differently from interactive on Google AI Ultra; is it permitted; wrapping/ACP policy.	YES. manirajc, a collaborator on the google-antigravity/antigravity-cli repo (GitHub label), answered 2026-11-06 (https://github.com/google-antigravity/antigravity-cli/issues/1411#issuecomment-6681347816). [vendor]
B	google-antigravity/antigravity-cli#1392	2026-11-02	May an official AGY agent invoke `agy -p` through its own shell/hooks (cross-model delegation)?	No (0 comments).
C	google-antigravity/antigravity-cli#1493	2026-11-10	Is `agy acp` from a non-editor ACP client permitted under the Additional Terms? Quotes Additional Terms s6 and the FAQ.	No (0 comments).
D	anthropics/claude-code#97074 https://github.com/anthropics/claude-code/issues/97074	2026-09-25 (older than 14 days, but the root of #1 and #2)	`claude -p` (entrypoint sdk-cli) uses ~1.8x more of the 5-hour window per token than interactive (entrypoint cli). 73 comments; no MEMBER/OWNER/COLLABORATOR comment found.	No.
E	anthropics/claude-code#106375 https://github.com/anthropics/claude-code/issues/106375	2026-10-31	Opus 5.5 usage depletes rapidly in SDK sessions after Oct 28 fixes.	No Anthropic reply; a user measurement (see 3).
F	anthropics/claude-code#106413 https://github.com/anthropics/claude-code/issues/106413	2026-10-31	`claude -p` on a subscription: claude.ai connectors added ~426k input tokens per call; `--bare` needs an API key.	No staff reply; community found the cause (see 5).
G	anthropics/claude-code#105540	2026-10-28	Claude Code silently bills to an ANTHROPIC_API_KEY env var set for another app instead of the subscription. Title only read; body not verified.	not checked

2. Per-tool table

| | Claude Code / Agent SDK | Codex CLI / exec | Gemini CLI / Antigravity CLI | GitHub Copilot CLI / cloud agent | | ----------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | (a) Scripted use metered differently from interactive? | Vendor does not say. Help article only says `claude -p` and Agent SDK "still draw from your subscription limits". Users measured ~1.8x to ~1.95x faster 5-hour drain for sdk-cli/sdk-ts vs cli; weekly not visibly different at 1-point resolution. [measured: qvarker #97074; douflou in #106375, 2026-11-02/03, Max plan, 12 sessions/pass; the meter followed the declared entrypoint (TUI with CLAUDE_CODE_ENTRYPOINT=sdk-cli metered like SDK)]. [opinion: Theo Browne video, https://youtu.be/y5TEATUDDfc, 12:00-13:39] | Vendor does not say. Docs read do not describe any mode-specific limit. Codex discussion 59674 asks exactly this, unanswered. | Antigravity: vendor says no. "Interactive agy, agy -p, and agy acp share the exact same 5-hour and weekly quota buckets and per-token weights." [vendor: manirajc, a collaborator on the google-antigravity/antigravity-cli repo (GitHub label), antigravity-cli#1411, 2026-11-06]. Cold prompt cache per new `agy -p` process makes one-shot runs use more quota per turn (same comment). [measured: jyongchul #1411, 0.98x per token on Flash, 2.5-7.2x per turn from cold cache]. Gemini CLI: quota page lists limits by auth method (AI Pro 1,500, Ultra 2,000 requests/day) and does not mention headless. | Copilot CLI: "Each prompt to Copilot CLI uses one premium request with the default model." Cloud agent: "one premium request per session, multiplied by the model's rate" (docs.github.com copilot-requests page). That page does not compare interactive vs automated. NOTE: the cloud-agent page now mentions "AI credits"; billing wording may be mid-change = not verified which unit applies today. | | (b) Vendor text on automated / headless / CI / third-party use of a consumer subscription | See section 3. Consumer Terms s3 bars bots/scripts "except when you are accessing our Services via an Anthropic API Key or where we otherwise explicitly permit it". Docs explicitly document subscription OAuth token for CI. | Docs: API key is "the recommended default for automation"; ChatGPT-managed login allowed "on trusted runners". OpenAI Terms of Use page returned HTTP 403 = not verified. | Antigravity FAQ: `agy -p` "is intended for direct scripting and CI workflows"; wrapping in third-party agents "not supported under consumer plans". Gemini CLI terms page: third-party software prohibition; does not address scripts/headless. | Docs describe programmatic mode "designed for scripts and automation". No subscription-terms statement found. | | (c) Documented CI route | `ANTHROPIC_API_KEY` (always used in -p when present), or `CLAUDE_CODE_OAUTH_TOKEN` from `claude setup-token` ("for CI pipelines and scripts"; one-year token; Pro/Max/Ultra/Team/Enterprise). GitHub Action inputs `anthropic_api_key` / `claude_code_oauth_token`. `--bare` ignores OAuth and needs the API key or apiKeyHelper. | `CODEX_API_KEY=<key> codex exec ...` for one invocation; or ChatGPT-managed `~/.codex/auth.json` on trusted private runners; Codex GitHub Action recommended. | Antigravity: `agy -p` uses cached credentials from a prior interactive login; unauthenticated CI run exits "authentication required". Google's own advice for unattended/third-party agents: AI Studio key `GEMINI_API_KEY` or Vertex AI (collaborator comment). Gemini CLI headless doc page did not cover auth = not verified for Gemini CLI. | `copilot -p PROMPT`, `--no-ask-user`; env `COPILOT_GITHUB_TOKEN`; PAT with "Copilot Requests" permission. | | (d) Enforcement events stated by the vendor | Only generic: "Anthropic reserves the right to take measures to enforce these restrictions and may do so without prior notice." No specific event found on vendor pages. Earlier 2026 events are in secondary press only = not verified. | None found. | Antigravity FAQ: third-party use "can result in suspension or termination of your account". Gemini CLI terms page: "may be grounds for suspension or termination". No named event. | None found. |

3. Quoted terms, URLs, read dates

All read 2026-11-11.

Anthropic

Help article "Use the Claude Agent SDK with your Claude plan", https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan (page shows "Updated over a week ago"). Update October 7, 2026: "Claude Max, Ultra, and Team plans now include monthly API credits." They cover "the Claude Agent SDK and `claude -p`" "when you run them with an API key from your linked Claude Console organization." "Nothing changes for your plan: when you're signed in with your Claude plan, the Claude Agent SDK and `claude -p` still draw from your subscription limits." Update June 15, 2026: "We've paused the previously-announced changes to Claude Agent SDK usage. For now, nothing has changed: Claude Agent SDK, `claude -p`, and third-party app usage still draw from your subscription limits." The article says nothing about a different rate. This is the sentence all the Anthropic threads cite.
Legal and compliance, https://code.claude.com/docs/en/legal-and-compliance: "Advertised usage limits for Pro, Max, and Ultra plans assume ordinary, individual usage of Claude Code and the Agent SDK." "OAuth authentication is intended exclusively for purchasers of Claude Free, Pro, Max, Ultra, Team, and Enterprise subscription plans and is designed to support ordinary use of Claude Code and other native Anthropic applications." "Developers building products or services that interact with Claude's capabilities, including those using the Agent SDK, should use API key authentication through Claude Console or a supported cloud provider. Anthropic does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, Max, or Ultra plan credentials on behalf of their users." Also: "Nor does it prevent an end user from signing in to the unmodified Claude Code binary with their own Claude subscription". "Anthropic reserves the right to take measures to enforce these restrictions and may do so without prior notice."
Consumer Terms, https://www.anthropic.com/legal/consumer-terms, effective Oct 8, 2025 per page. Section 3 item 7: "Except when you are accessing our Services via an Anthropic API Key or where we otherwise explicitly permit it," "to access the Services through automated or non-human means, whether through a bot, script, or otherwise."
Run Claude Code programmatically, https://code.claude.com/docs/en/headless: "It's available as a CLI for scripts and CI/CD". "Set `ANTHROPIC_API_KEY` before running it, because bare mode doesn't use your subscription login". "In bare mode, Claude Code never reads OAuth credentials or the system keychain." "`--bare` is the recommended mode for scripted and SDK calls, and will become the default for `-p` in a future release." (So the recommended scripted mode needs an API key.)
Authentication, https://code.claude.com/docs/en/authentication: precedence list item 3 "In non-interactive mode (`-p`), the key is always used when present."; item 5 "`CLAUDE_CODE_OAUTH_TOKEN` ... Use this for CI pipelines and scripts where browser login isn't available."; "generate a one-year OAuth token with `claude setup-token`"; "This token authenticates with your Claude subscription and requires a Pro, Max, Ultra, Team, or Enterprise plan." Gotcha in the same page: an `ANTHROPIC_API_KEY` set in the environment is used once approved even if you have a subscription ("Run `unset ANTHROPIC_API_KEY` to fall back to your subscription").
GitHub Actions, https://code.claude.com/docs/en/github-actions: "`CLAUDE_CODE_OAUTH_TOKEN`: an OAuth token that authenticates with your Claude subscription, available on Pro, Max, Ultra, Team, and Enterprise plans." "If you authenticate with an OAuth token, runs use your Claude subscription instead of API billing." "For a secret shared across repositories, authenticate with an API key ... rather than an OAuth token, since an OAuth token is tied to the subscription of the person who ran `claude setup-token`."

OpenAI

Codex authentication, https://developers.openai.com/codex/auth (308 redirect to https://learn.chatgpt.com/docs/auth): "Sign in with ChatGPT for subscription access." "Sign in with an API key for usage-based access." "Use API key authentication for programmatic Codex CLI workflows, such as CI/CD jobs." "API keys are still the recommended default for automation." "Access tokens are intended for trusted scripts, schedulers, and private CI runners." "When you sign in with an API key, Codex uses standard API pricing instead of included ChatGPT plan credits." "Treat `~/.codex/auth.json` like a password". The page says nothing on usage limits for `codex exec`.
Non-interactive mode, https://learn.chatgpt.com/docs/non-interactive-mode: `codex exec` "reuses saved CLI authentication by default"; `CODEX_API_KEY=<api-key> codex exec --json "..."`; "Do not set `OPENAI_API_KEY` or `CODEX_API_KEY` as a job-level environment variable."; ChatGPT-managed login in CI is for teams "using ChatGPT-managed Codex access on trusted runners" or who need "ChatGPT/Codex rate limits instead of API key usage"; API keys are "the right default for automation"; "Do not use this workflow for public or open-source repositories".
OpenAI help center Codex plan-usage article (help.openai.com article 11369540) returned HTTP 403: not verified. OpenAI Terms of Use returned HTTP 403: not verified. The only text seen is quoted by discussion 59674's author from the Terms: "Interfere with or disrupt our Services, including circumvent any rate limits or restrictions or bypass any protective measures or safety mitigations we put on our Services." [secondhand, not verified against the live page]

Google

Antigravity FAQ, https://antigravity.google/docs/faq: "Antigravity CLI headless mode (`agy -p`) is intended for direct scripting and CI workflows." Wrapping `agy` in third-party tools or automating account rotation "is not supported under consumer plans." "Using third-party software, tools, or services to access Antigravity is a violation of our Terms of Service ... Such actions can result in suspension or termination of your account." "To use a third-party coding agent with Gemini, we recommend using a Gemini Enterprise or Google AI Studio API key."
Antigravity headless docs, https://antigravity.google/docs/cli/headless: "Headless mode uses your cached credentials. Authenticate once with an interactive `agy` session first."; in CI an unauthenticated run "exits with an `authentication required` error instead of hanging."
manirajc, a collaborator on the google-antigravity/antigravity-cli repo (GitHub label), antigravity-cli#1411, 2026-11-06 [vendor]: "Direct scripting and CI workflows using agy -p on a single personal account are officially supported under consumer plans"; "Interactive agy, agy -p, and agy acp share the exact same 5-hour and weekly quota buckets and per-token weights."; "Cached input tokens ... carry a much lower quota weight than uncached input tokens"; "wrapping agy -p inside third-party autonomous agents/queues, or using multi-account rotation/pooling hubs (such as CLIProxyAPI) violates the Terms of Service. For third-party agents or unattended automation, use a Google AI Studio API key (GEMINI_API_KEY) or Vertex AI."; Google "does not grant private exemptions" from the Terms. Google said it is updating the Headless, CLI Reference and /usage docs to explain cold vs warm cache.
Gemini CLI terms and privacy, https://github.com/google-gemini/gemini-cli/blob/main/docs/resources/tos-privacy.md: for AI Pro/Ultra login, the page lists three terms: "Google Terms of Service", "Google One Additional Terms of Service" and "Google Privacy Policy"; using "third-party software, tools, or services" is "a violation of applicable terms and policies" with OpenClaw-on-Gemini-CLI-OAuth as the example; "Such actions may be grounds for suspension or termination of your account." The page does not address scripts, automation or headless use. This is the gap behind discussion 30030.
Gemini CLI quota page, https://github.com/google-gemini/gemini-cli/blob/main/docs/resources/quota-and-pricing.md: Google AI Pro 1,500 and Ultra 2,000 requests per user per day; free API key 250 (Flash only); "Quotas and pricing are based on a fixed price subscription." Does not mention headless.

GitHub

Copilot requests, https://docs.github.com/en/copilot/concepts/billing/copilot-requests: "Each prompt to Copilot CLI uses one premium request with the default model." "Copilot cloud agent uses one premium request per session, multiplied by the model's rate." For agentic features "only the prompts you send count as premium requests."
Automate with Actions, https://docs.github.com/en/copilot/how-tos/copilot-cli/automate-with-actions: "Programmatic mode is designed for scripts and automation and lets you pass a prompt non-interactively." `copilot -p PROMPT [OPTIONS]`; PAT with "Copilot Requests" permission; `COPILOT_GITHUB_TOKEN`.
About cloud agent, https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent: "cloud and GitHub Actions compute draw on the same monthly GitHub account allowance, separate from AI credits."

Community measurement and opinion (not vendor)

[measured: qvarker, https://github.com/anthropics/claude-code/issues/97074, 2026-09-25] Max plan, Claude Code 2.1.282, Opus 5, same workload: headless ~$1.6 API-equivalent per 1% of 5h window vs interactive ~$3.0.
[measured: douflou, comment in https://github.com/anthropics/claude-code/issues/106375, 2026-11-03] Max, 2.1.316, Opus 5.5: TUI 4.9 five-hour points per $10, `claude -p` and Agent SDK 9.6, TUI with CLAUDE_CODE_ENTRYPOINT=sdk-cli 9.9; ~1.95x; weekly not distinguishable at 1-point resolution; single-pass caveat on the entrypoint row.
[measured: tomahawkeer, https://github.com/anthropics/claude-code/issues/106413, 2026-10-31] 429,388 input tokens per `claude -p --tools ""` call on a subscription login; most from claude.ai connector tool definitions loaded upfront because `--tools ""` removes ToolSearch; 2,854 with `--strict-mcp-config`. Corrected by iamnikacode (plain `claude -p` ~60k). [opinion: iamnikacode] cites the MCP docs sentence "If you deny `ToolSearch` or leave it out of `--tools`, MCP tools load upfront instead." Workaround env var `ENABLE_CLAUDEAI_MCP_SERVERS=false` or `"disableClaudeAiConnectors": true` stated by iamnikacode with a link to code.claude.com/docs/en/mcp#disable-claude-ai-connectors (the reporter repeated it); the MCP docs page was not opened.
[opinion: Theo Browne] T3 Code author says Anthropic informally allowed a custom CLAUDE_CODE_ENTRYPOINT for T3 Code (as paraphrased in #107106). Not vendor-confirmed.

4. Gaps (what the vendors do not say)

Anthropic does not say whether `claude -p` / Agent SDK use is counted at a different rate than interactive use. The only vendor sentence is "still draw from your subscription limits". Two independent-looking measurements (qvarker, douflou) show ~1.8-1.95x on the 5-hour window; no Anthropic staff comment found on #97074 (73 comments), #107106, #107134, #106375.
Anthropic does not say whether setting `CLAUDE_CODE_ENTRYPOINT` yourself is allowed. It is not in the documented env-var list (per #107106; the env-vars page was not re-checked).
Anthropic's Consumer Terms bar "automated or non-human means ... script" unless "we otherwise explicitly permit it". The docs do document `claude setup-token` for CI/scripts, but none of the pages read says in so many words that this counts as explicit permission.
Anthropic's legal page says limits "assume ordinary, individual usage". "Ordinary" is undefined; no number of unattended runs or hours is given.
Mods/hooks sending background requests on Pro/Max (#106161): the vendor does not say.
Cloud routines and `claude -p` subprocesses: which meter they bill (#106850): the vendor does not say in any page read. Behaviour at a usage limit in headless mode (#108211): no documented exit-on-limit flag found on the headless page; the full CLI reference was not read, so treat "no flag" as not verified.
OpenAI does not say (in the pages that could be loaded) whether `codex exec` is metered like the interactive TUI on a ChatGPT plan. It recommends API keys for automation and allows ChatGPT login on trusted runners. OpenAI help center and Terms returned 403 = not verified.
Google: for Gemini CLI itself, the terms page prohibits "third-party software" access and does not address your own script calling `gemini -p`. For Antigravity CLI, the collaborator comment (manirajc, a collaborator on the google-antigravity/antigravity-cli repo) and the FAQ say `agy -p` scripting/CI is supported on a single personal account. Whether that carries over to Gemini CLI is not stated.
GitHub: no statement found on Copilot subscription seats used by unattended Copilot CLI beyond billing units; the docs billing wording is shifting toward "AI credits" (not verified which applies today).
No vendor-stated enforcement event (named wave of bans or blocks) found for any of the four. Secondary press reports exist but were not verified.
Search sweep for other recent threads was cut short by a GitHub API rate limit; openai/codex, gemini-cli and copilot-cli repos were only partly swept.

5. Answer per question (single fact, or "not answerable")

#107106 (rate difference / ENTRYPOINT / approved route): Rate: not answerable from a vendor source ([vendor] says only "still draw from your subscription limits"; [measured: douflou, #106375] 9.6 vs 4.9 five-hour points per $10, ~1.95x). Approved unattended route: [vendor] https://code.claude.com/docs/en/authentication says `CLAUDE_CODE_OAUTH_TOKEN` from `claude setup-token` is "for CI pipelines and scripts", and an API key is always used in `-p` when present. ENTRYPOINT: not answerable.
#107134 (is `-p` permitted; metered at 1.8x?): `-p` is documented for scripts and CI, and the subscription token is documented for CI; the Consumer Terms bar bots/scripts unless an API key is used or Anthropic "otherwise explicitly permit[s] it". The vendor does not say the CI docs are that permission. Rate: not answerable from a vendor source.
#108211 (silent wait at usage limit): not answerable from a vendor source. Only [opinion: fortitude-omnis, vijaym2k6] workarounds (pid file, `timeout`).
#106850 (routine `claude -p` bills which meter): not answerable.
#108851 (non-interactive limit reset): not answerable.
#106161 (mods background requests): not answerable. Only the general rule can be quoted: Consumer Terms s3 item 7 and "assume ordinary, individual usage".
Codex discussion 59674 (quota parity TUI/exec/app-server): not answerable. Nearest vendor fact: https://learn.chatgpt.com/docs/non-interactive-mode says ChatGPT-managed login in CI uses "ChatGPT/Codex rate limits instead of API key usage" and API keys are "the right default for automation".
Gemini CLI discussion 30030 (is `gemini -p` on AI Pro allowed): not answerable for Gemini CLI specifically (terms page silent on scripts). Adjacent [vendor] fact for Antigravity CLI: manirajc, a collaborator on the google-antigravity/antigravity-cli repo, 2026-11-06, "Direct scripting and CI workflows using agy -p on a single personal account are officially supported under consumer plans"; for third-party agents or unattended automation Google says use an AI Studio key or Vertex.

Extra answered by a vendor: Antigravity #1411: "Interactive agy, agy -p, and agy acp share the exact same 5-hour and weekly quota buckets and per-token weights" (cold cache is why one-shot runs cost more per turn).