LibraOS End User Agreement

1. Acceptance, Parties, and Authority

1.1. Parties. This End User Agreement (the "Agreement") is a contract between you and Nebula Nova Inc., a Delaware corporation ("Nebula Nova", "we", or "us"), for the LibraOS Software and Services.

1.2. Acceptance. Before your first use, your first Gateway request on each route (Section 5.3), and any material change (Section 13.4), we will show you the terms conspicuously and ask you to accept by an unchecked box or "I agree" button. For AgentD, we ask when you set it up with agentd init, before it runs any work, and again in agentd activate if you have not accepted yet; AgentD records your acceptance (agreement version, time and user) on your machine. AgentD sends acceptances to LibraOS only from an install linked to your account (Section 4.6): the acceptance record saved at the time of registration under Section 6.4(a), including when reporting is off and you request feedback or a log-bundle upload, and later acceptances, including when reporting is off. An install that is not linked keeps its acceptance on your machine and sends none. Each re-acceptance report contains the agreement version and SHA-256 digest, the saved acceptance time and user, acceptance method (prompt or flag), and the AgentD version, operating system and processor type, authenticated with the installation credential. If needed it first attempts registration as for feedback. If offline or a request fails, AgentD keeps the acceptance pending and retries on the next activation, init and daily run; doctor warns only when a registered install has an acceptance report pending retry. An install without an installation credential passes with a note that acceptance is kept locally and not recorded by LibraOS. Doctor sends no acceptance report, does not register, and writes no acceptance, pending record or registration state. Doctor still fetches each known checkout’s origin (updating remote-tracking refs) and probes the Runtime, gateway, cloud knowledge and GitHub. When a saved acceptance is for another version, agentd init and agentd activate ask again; agentd doctor warns without failing or recording acceptance of the new version. For the Services, we record each acceptance with a timestamp, user, organization, method, and the versions of this Agreement and Gateway Provider Terms accepted.

1.3. Authority. If you accept for an organization, you represent that you may bind it, and "you" means it. A "Consumer" is an individual acting mainly outside their trade, business, or profession.

1.4. Related Documents and Precedence. This Agreement incorporates the Website Terms (https://libraos.com/terms), the terms and prices on the Credits & billing page at https://console.libraos.com/credits (the "Billing Terms"), and the Section 5.3 obligations. Our Privacy Policy (https://libraos.com/privacy) governs personal data. On conflict, in order: (a) an open source license, for its component; (b) for personal data, the Privacy Policy, subject to mandatory law, but the commitment more protective of the individual applies; (c) the Section 5.3 obligations, for that route; (d) this Agreement; (e) the Billing Terms, for pricing and payment; (f) the Website Terms.

1.5. Age. You must be at least 18 and the age of majority where you live.

2. Definitions

"AI Provider" means Anthropic, OpenAI, or another model provider we name under Section 5.3.

"Authorized User" means an individual an organization permits to use the Software on its behalf.

"Credentials" means passwords, nk_ and lk_ keys, GitHub tokens, AI Provider keys and sign-ins, and other secrets used with the Software.

"Input" means material submitted to the Software or to an AI Provider through it. "Output" means content an AI Provider generates in response. "User Content" means Inputs, Outputs, and other content you store in the Services.

"Services" means the hosted LibraOS Cloud control plane at app.libraos.com and console.libraos.com, including workspaces, subdomains, keys, Chat, Builder, credits, Token Plans, and the model API gateway.

"Software" means the Services, AgentD, self-hosted LibraOS, Libra Desk, and LibraOS SDKs, installers, and documentation.

"Worker" means a Claude Code or Codex session that AgentD starts and coordinates on your machine.

3. License Grant and Restrictions

3.1. License. Subject to this Agreement, we grant you a limited, non-exclusive, non-transferable, non-sublicensable license to use the Software's proprietary components and access the Services for your internal business or personal purposes. An organization may let its Authorized Users use it on its behalf; that is not a transfer or sublicense, and the organization is responsible for their compliance. We may suspend or end the license only under Section 12. AgentD Core (the AgentD tool running on your machines, without Cloud Workers or Gateway Access) is provided free of charge. Your license to each AgentD Core version you have obtained is perpetual and irrevocable except under Section 12.3(a), and stays in force through any assignment, change of control, change of these terms, or end of AgentD. We, or a successor, may offer later versions under different terms, and you choose whether to accept them; Section 13.4 applies to any change.

3.2. Open Source. Open source components are governed by their own licenses, listed in their notice files, which this Agreement does not restrict.

3.3. Restrictions. Except as an open source license or law permits, you will not: (a) copy or modify proprietary components, except as needed to use them; (b) sell, rent, sublicense, or transfer the Software, except under Section 3.1; (c) reverse engineer or decompile proprietary components; (d) use our confidential information, or copy our proprietary code, to build a competing product; (e) remove proprietary notices; (f) bypass security, metering, or access controls; or (g) share, pool, or resell lk_ key access outside your account. Lawful benchmarking, interoperability work, and independent development are not restricted.

4. Accounts and Credentials

4.1. Registration. You need a verified email address and accurate, current account information.

4.2. Responsibility. You are responsible for activity under your account or Credentials that you or your Authorized Users authorize or that your failure to protect them reasonably allows, but not for unauthorized activity our security failure causes. Report suspected misuse to the security contact in Section 14.

4.3. Keys. For nk_ workspace keys, we store only the key id, never the secret; a lost nk_ key is re-minted, not recovered. For lk_ model API keys, [lk_ key handling]. Revoke and replace any exposed key.

4.4. Connected Accounts. You may connect AI Provider sign-ins, API keys, and GitHub accounts you are entitled to use. For local Workers, these are held by configured tools such as Claude Code, Codex, opencode, git, or the GitHub CLI. AgentD also uses configured model and integration Credentials to authenticate the requests described in Section 6.4, including hosted operations; it does not include those Credentials in automatic usage reports. If you store a Credential in the Services, [stored-credential handling]. Grant tokens minimal scopes.

4.5. Personal Benefits. Invite codes, activation codes, and credits are personal and may not be sold, transferred, or shared outside your account.

4.6. Account Linking. AgentD Core works without a LibraOS account. To use features that need your account, such as Cloud Workers, the hosted knowledge base in Section 6.4(e), and sending feedback or diagnostic log bundles, you link an install with agentd activate and the activation code from your Developer Console account. The code is signed by us and verified on your machine; it has no scheduled expiry, is not checked with us again, and needs no network connection to keep working (AgentD may re-verify the saved code locally for integrity). You may use the one code on every machine that you, or your organization's Authorized Users, use for your account; Section 4.5 still applies, so do not pass it outside your account. The code by itself grants no access to your account, keys, credits or Cloud Workers, and we do not disable installed copies of AgentD Core remotely. Ending this Agreement under Section 12 ends your Console access; Section 12.4 says what happens to the AgentD Core versions you have.

5. Third-Party AI Providers

5.1. Access Modes. AgentD supplies no model access; Workers call their models as their client is configured; AgentD's model-backed features use the configured model endpoint. LibraOS Cloud supplies model access only as Gateway Access. You use either:

(a) Direct Access: your own Claude or ChatGPT plan, or Anthropic or OpenAI API key, configured separately in the client. Requests go to the AI Provider, not through us, and it bills you; or

(b) Gateway Access: an lk_ model API key from the Developer Console. Requests go through our gateway to the AI Provider and are paid from your credits or Token Plan under the Billing Terms.

5.2. Direct Access Terms. Your own agreement with the AI Provider, including its usage policies (https://www.anthropic.com/legal, https://openai.com/policies), governs Direct Access and its handling of your data. We are not a party to it.

5.3. Gateway Provider Terms. Each Gateway route's "Gateway Provider Terms" are: (a) Anthropic (Claude): the Commercial Terms of Service (https://www.anthropic.com/legal/commercial-terms) and Usage Policy (https://www.anthropic.com/legal/aup); (b) OpenAI (GPT and Codex): the Services Agreement (https://openai.com/policies/services-agreement), Service Terms (https://openai.com/policies/service-terms), and Usage Policies (https://openai.com/policies/usage-policies). Only these obligations are incorporated, in the version you accepted under Section 1.2, and you owe them to us, not the provider, unless it contracts with you separately: (i) comply with the route's usage policies; and (ii) follow the restrictions its terms place on end users, including any bar on using Outputs to develop competing models. You must accept changed Gateway Provider Terms before your next request on that route, or, if a change materially harms you, may stop Gateway Access with a refund under Section 12.4. New providers are named with their terms before you can select them. We may block requests we reasonably believe violate these obligations.

5.4. Client Software. Claude Code (Anthropic) and Codex (OpenAI) are licensed to you by their vendors; AgentD neither includes nor licenses them.

5.5. Provider Decisions. AI Providers may change or withdraw models, prices, or policies, or suspend your provider account. We are not responsible for those decisions where they are outside our reasonable control, but are responsible to the extent our routing, billing, failure to give notice we owe, or breach caused your loss.

6. Inputs, Outputs, and User Content

6.1. Ownership. As between us, you own your User Content, and we assign to you any rights we have in Outputs. Similar Outputs may be generated for others.

6.2. License to Us. For content you store in the Services or send through Gateway Access, you grant us a non-exclusive, royalty-free license to host, process, and transmit it only to provide, secure, and support the Services for you, ending for each item at its deletion under Section 6.6, except for backups within the backup period and data under a legal hold or open dispute. (a) Hosting subprocessors, listed at [subprocessor list URL], process it only for us under written terms requiring confidentiality, security, and deletion as Section 6.6 provides. (b) Gateway AI Providers are not our subprocessors; their Gateway Provider Terms, which we do not control, govern their processing and retention of Gateway content. You represent that you have the rights and consents to submit your User Content.

6.3. Training. Nebula Nova does not use User Content to train any AI model. For Gateway Access, the AI Providers' commercial terms, as of the effective date, exclude training on API inputs and outputs by default, and we have not opted in [confirm per route]. For Direct Access, your agreement and settings with the AI Provider govern training.

6.4. What Leaves Your Machine. A Worker, or AgentD's model-backed features (including Chat, voice transcription, ask and copilots), may send its AI Provider prompts, conversation and retrieved context, audio and images you submit, any file or command output it can read, and large parts of a repository, depending on the client, its configuration, the permissions you enable, and your instructions. With Direct Access it goes to the AI Provider and we cannot see it. With Gateway Access our gateway routes and meters it. AgentD also requests model lists and prices from configured model endpoints, authenticated with the configured key. AgentD itself makes the following service requests, only as described; requests to your own or third-party services go to those services, not necessarily to us. (a) Registration, once per linked machine, when reporting is on or you accept the agreement or request feedback or a log-bundle upload, and a network is available: your activation code, an installation id, the AgentD version, operating system and processor type, and your acceptance record from Section 1.2; we answer with an installation credential that AgentD keeps on that machine and uses for (b), (c) and (d), and the activation code is not sent again. Re-acceptance reports under Section 1.2 use that credential and send no activation code, paths or repository names; we record these acceptances with registration events under Section 6.6(h). (b) A daily usage report: the installation id, version, operating system, processor type, the UTC day, and that day's counts of handoffs, workers started, finished and failed, sessions and users, tied to your account. (c) Feedback you choose to send from AgentD: your text, a rating and category if you give them, the AgentD version, operating system, processor type and feedback source, and, only if you include it, a diagnostic report. (d) Diagnostic log bundles you choose to send with agentd logs export --send: AgentD first writes the bundle on your machine, lists the files in it with their sizes, the number of redactions and the Console address, and sends it only when you confirm at the prompt or have given --yes on the command line; without a terminal and without --yes it sends nothing. A bundle can hold, for the tasks you name or, unless you name only Codex sessions, every task updated in the chosen period (24 hours unless you choose another), their task records, worker logs and worker transcripts, any Codex session you name, token use and timing, your configuration with recognised Credentials and paths under .secrets/ replaced, and a diagnostic report of the AgentD version, operating system, processor type and local checks; with it go your optional note, the AgentD version, operating system and processor type, and the archive's SHA-256 digest and file name, sent with the installation credential (or a legacy download token if registration fails). The optional note is sent as given, without the bundle's redaction filter. AgentD applies a credential-pattern filter to exported record, log, transcript and configuration contents, replacing matches with "[redacted]", and does not include files under .secrets/ in the bundle (local checks and authentication can read credential files), but a bundle can contain code, prompts, file paths, repository names and Outputs from those tasks. (e) Knowledge you choose to sync with agentd cloud connect: AgentD first shows how many records of each kind it will upload (pull requests, issues, runs, handoffs and session summaries), their approximate total size and the workspace address, and connects only when you confirm at the prompt or have given --yes; without a terminal it refuses unless you give --yes. It uploads those records from the ledger on that machine to your LibraOS workspace's knowledge base, with the workspace key as the Authorization header, so that your agents can search them, and keeps that copy in sync as the records change. A configured local Runtime takes priority, so cloud uploads and searches pause while it is configured; AGENTD_NO_KB also pauses uploads. When the cloud knowledge base is active, AgentD also sends it search queries and agentd ask questions to retrieve records. The ledger refuses fields that look like Credentials when they are written, but handoffs and session summaries can contain code, file paths, repository names, decisions and excerpts of prompts and replies; separate message, prompt and decision records, full transcripts and files under .secrets/ are not uploaded. Uploading stops when you run agentd cloud disconnect, and agentd cloud disconnect --delete (or answering yes when it asks) also deletes the hosted copy. (f) Hosted tasks you request with agentd handoff offer --cloud: AgentD sends the task id, goal, repository name and budget to the configured Job API, authenticated with your account key, and requests task status, results, cancellation and account balance as needed. If you choose --model-access byok, AgentD also sends the Anthropic or OpenAI API key you name to the Job API with the task, and again if the Job API asks for it after a restart; the Job API keeps it in memory only, passes it to that task's worker environment, does not store or log it, and discards it when the task ends. The task's model requests then go from the worker to that AI Provider as Direct Access (Section 5.1(a)), and the AI Provider bills you. (g) Release checks and downloads: with reporting on, the daily run of a linked install checks for a newer release; agentd upgrade also requests release files and checksums when you run it. These requests identify the requested release, operating system and processor type in the URL. Downloads may follow redirects to the release mirror. (h) Configured Runtime: when you configure a Runtime, AgentD sends its login email and password or token to that address, probes health and capabilities, and sends ledger records, rules documents, search queries, agent definitions and questions to its knowledge and agent APIs. The recall service also uploads transcript stretches for storage and model extraction; these can contain code, prompts and Outputs, unlike the cloud knowledge mirror in (e). A Runtime may be on another host; no cloud-connect confirmation is required for these configured Runtime operations. (i) Connected services: configured Git repositories and GitHub receive repository identifiers, Git traffic and pull-request or issue queries, and changes you request through git or the GitHub CLI using those tools' credentials. Linear issue sync sends team identifiers and pagination queries with the configured API key. agentd runtime up may also pull container images from the configured image registries. Configured notifications send notification titles (including excerpts of task goals or approval requests), item ids and dashboard links to Telegram or Gmail and the recipients; Telegram requests use the bot token and chat ids and poll and answer bot interactions, and Gmail uses the configured sender and app password to send mail, including invitation links you request. (j) Hosting operations: if you run AgentD's Job API or cloud worker, it requests account authentication, credit holds and settlement, model-key issuance and revocation, usage and balances from Console using service credentials and account/job identifiers; authentication also sends the account key. It requests repository installation tokens from GitHub and revokes them. Cloud workers exchange leases, repository and model credentials, status, results and cost with the Job API; they send repository traffic to the repository host, and, when a runner is configured, commands, environment variables and repository credentials to that runner to execute the job and push its branch. (k) Commands you configure: AgentD also runs verification checks and benchmark commands you configure, as well as configured watchdog restart commands. Depending on their permissions and configuration, these can read local files and environment variables and make network requests to other services; you control the destinations, data and Credentials those commands use. External tools AgentD launches, including worker clients and login-status probes, may also contact their vendors under those tools' own settings and terms. These commands and configured integrations operate independently of automatic reporting. Third-party services apply their own retention terms; LibraOS-hosted content follows Section 6.6, and you control storage in self-hosted Runtimes and runners. Automatic registration and daily reports do not include code, prompts, file paths, repository names or Outputs; registration does send your activation code in (a). The bundle redaction described in (d) is not a filter on feedback text, upload notes, knowledge queries, hosted task goals or model requests; do not include Credentials in them. When we receive any of these we also see the sending IP address and any client version information it carries, and we derive an approximate location (country and city) from the IP address. You may turn automatic reporting off at any time with agentd config set telemetry.enabled false or the DO_NOT_TRACK environment variable: that stops registration and daily reports, does not stop feedback, log bundles or knowledge sync you start yourself, or acceptance reports (sending feedback or a log bundle registers the machine under (a) first if it is not registered yet), and does not change how the Software works. An install that is not linked to your account under Section 4.6 sends no registration, usage reports, acceptance reports or automatic release checks to LibraOS, and refuses feedback, log-bundle uploads, knowledge sync and hosted tasks until you link it; the requests in (h) to (k) that you configure still happen. Reporting is on by default for a linked install, and AgentD tells you so when you link it. We may revoke an installation credential, which stops that machine's reports and requests using that credential, including feedback and log-bundle uploads. The Privacy Policy describes this data too.

6.5. Usage Metadata. We collect usage metadata (such as model, token counts, timestamps, and key ids) for billing, usage reports, security, and improving the Services, and do not sell it or your User Content. We may keep de-identified metadata that cannot reasonably be linked to any individual, and will not re-identify it.

6.6. Retention and Deletion. We keep: (a) stored content until you delete it or your account is deleted; (b) Gateway request and response content for 90 days; (c) usage metadata and billing records for 90 days, or longer where tax law requires; (d) account records for 90 days after account closure; (e) security logs for 90 days; (f) support messages for 90 days; and (g) the record of your AgentD activation code for as long as your account has AgentD access, so that the Console can show you the same code; (h) AgentD registration and usage-report events, including the IP addresses and locations received with them, for 12 months, deleted on a schedule; (i) feedback you send from AgentD or the Console for 24 months, or until you ask us to delete it; (j) diagnostic log bundles you send from AgentD for 30 days, then deleted, or earlier if you ask us; and (k) knowledge you sync with agentd cloud connect, as stored content under (a), until you delete it with agentd cloud disconnect --delete, delete your workspace, or your account is deleted. Backups are erased, and hosting subprocessors delete their copies, within 30 days after our primary copy is deleted. We keep specific data longer only for a legal hold or open dispute. You may delete stored content yourself or through support, which confirms within 30 days. Gateway AI Providers retain Gateway content for the periods their Gateway Provider Terms set; Direct Access retention follows Section 5.2. You control self-hosted installations.

6.7. Feedback. We may use feedback you give without obligation to you, but not your confidential information or personal data within it.

7. Acceptable Use

7.1. You will not use, or let Workers, Authorized Users, or others use, the Software to: (a) violate law or third-party rights; (b) create or distribute malware, or commit fraud; (c) access systems, data, or workspaces without authorization; (d) change repositories or open pull requests without permission; (e) circumvent security, rate limits, credit checks, or safety measures; (f) violate applicable AI Provider terms or usage policies; or (g) extract model weights or proprietary system internals.

7.2. Enforcement. If we reasonably believe this Section is breached, we may act proportionately (such as blocking a request or suspending access) after notice and a chance to cure where practicable, and immediately only for a legal requirement, security risk, or risk to our AI Provider accounts. On request, someone who did not make the decision will review it within 14 days.

8. AI-Generated Output, Agent Risks, and Costs

8.1. Accuracy. Outputs may be wrong, incomplete, insecure, or out of date.

8.2. Review. Review Outputs before relying on them, and review every Worker change before merging or deploying it.

8.3. Worker Actions. Workers edit files, run commands, commit, and open pull requests under the permissions and instructions you give them. You are responsible for their actions within those limits, but not to the extent an action results from our Software defect, security failure, breach, or negligence, or our instructions beyond your configured permissions.

8.4. No Professional Advice. Outputs are not legal, medical, financial, tax, or other professional advice.

8.5. High-Risk Uses. You will not use Outputs where failure could cause death, injury, or severe damage (such as in critical infrastructure, medical devices, or weapons), or for decisions with legal or similarly significant effects on people (such as employment, credit, or housing) unless law and AI Provider terms permit it and a qualified person reviews each decision.

8.6. Costs. Loops, retries, and parallel Workers can raise Direct Access and Gateway Access usage sharply. Subject to Sections 4.2, 5.5, and 8.3, you are responsible for usage under your account and Credentials, including unexpected usage.

8.7. Billing. Purchased credits are prepaid and do not expire while your account is open; Section 12.4 governs them on termination. Free or promotional credits have no cash value. We will disclose any Token Plan auto-renewal before purchase, and you may cancel before the next renewal. Recurring-plan price increases need 30 days' notice and apply from your next renewal. Prices exclude taxes. If payment fails, we may pause Gateway Access until paid.

9. Intellectual Property and Trademarks

9.1. Ownership. We and our licensors own the Software and the LibraOS and Libra Desk names and logos.

9.2. Trademarks. You may not use our names or logos without written permission, except to refer accurately to LibraOS or Libra Desk without implying our endorsement.

9.3. Open Source. Section 3.2 governs open source components.

9.4. Copyright Complaints. Send Digital Millennium Copyright Act notices to our designated agent: [registered DMCA agent].

10. Warranty Disclaimer

10.1. TO THE EXTENT THE LAW ALLOWS, THE SOFTWARE AND OUTPUTS ARE PROVIDED "AS IS" AND "AS AVAILABLE", AND NEBULA NOVA DISCLAIMS ALL IMPLIED WARRANTIES, INCLUDING MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE, AND NON-INFRINGEMENT.

10.2. WE DO NOT WARRANT THAT THE SOFTWARE WILL BE UNINTERRUPTED, SECURE, OR ERROR-FREE, THAT OUTPUTS WILL BE ACCURATE, OR THAT ANY MODEL WILL REMAIN AVAILABLE. PRIVATE BETA AND PREVIEW FEATURES MAY CHANGE OR END.

10.3. Self-Hosting. You are responsible for the infrastructure you self-host the Software on.

10.4. Mandatory Warranties. This Section 10 does not exclude non-excludable statutory warranties; where law permits limiting their remedy, it is re-supply or refund, at our choice.

11. Limitation of Liability and Indemnity

11.1. Consumer Rights. Nothing in this Agreement limits rights that applicable law, including consumer protection law, makes non-waivable; Sections 10, 11.2, 11.3, and 11.5 apply only as far as that law allows.

11.2. Excluded Damages. TO THE EXTENT THE LAW ALLOWS, NEITHER PARTY IS LIABLE FOR INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, OR PUNITIVE DAMAGES, OR LOST PROFITS, REVENUE, OR GOODWILL.

11.3. Cap. TO THE EXTENT THE LAW ALLOWS, NEBULA NOVA'S TOTAL LIABILITY UNDER THIS AGREEMENT IS LIMITED TO THE GREATER OF THE AMOUNTS YOU PAID US IN THE 12 MONTHS BEFORE THE CLAIM AROSE AND USD $100. A CLAIM THAT WE CAUSED UNEXPECTED DIRECT ACCESS CHARGES IS SUBJECT TO THIS CAP, EXCEPT AS SECTION 11.4 PROVIDES.

11.4. Exceptions. Sections 11.2 and 11.3 do not apply to liability for death or personal injury caused by negligence, fraud, gross negligence, wilful misconduct, our breach of Section 6 or of our confidentiality or data-protection obligations, or Section 11.6.

11.5. Your Indemnity. If you are not a Consumer, you will defend and indemnify Nebula Nova against third-party claims to the extent caused by your or your Authorized Users' breach or your unlawful or infringing User Content.

11.6. Our IP Indemnity. If you are a "Business Customer" (a non-Consumer that paid us fees in the 12 months before the claim), we will defend you against third-party claims that our proprietary components, as provided and used under this Agreement, infringe a patent, copyright, or trademark or misappropriate a trade secret, and pay damages finally awarded and settlements we approve. Excluded are Outputs, User Content, open source components, AI Provider models and clients, changes we did not make, and combinations, but only where the claim would not have arisen without combining our components with something we did not provide, require, document, or reasonably anticipate. If use is or may be enjoined, we will at our expense secure continued use or make the component non-infringing with materially the same function, or else end the affected service and refund unused prepaid fees. This Section is your exclusive remedy for third-party infringement claims about the Software.

11.7. AI Providers. This Agreement does not release or limit any claim you have against an AI Provider.

11.8. Claim Procedure. For Sections 11.5 and 11.6, the defended party will give prompt written notice (delay excuses the other only to the extent prejudicial), let the defending party control the defense, and reasonably cooperate. The defending party reimburses reasonable out-of-pocket cooperation costs and any separate counsel a conflict of interest requires; otherwise the defended party may add its own counsel at its own cost. No settlement may admit fault for, or bind, the defended party without its consent, not unreasonably withheld.

12. Term, Termination, and Suspension

12.1. Term. This Agreement runs from your acceptance until terminated.

12.2. By You. You may terminate at any time by closing your account in the Developer Console.

12.3. By Us. We may suspend or terminate access, proportionately and with notice under Section 7.2, if: (a) you materially breach this Agreement (substantially breach Section 3.3, 7, or 13.3, or cause or risk significant harm to others) and fail to cure within 14 days after notice, or 30 days if cure reasonably needs longer and you pursue it diligently (incurable breaches have no cure period); (b) your use poses a security risk; (c) an AI Provider, payment processor, or law requires it; or (d) undisputed payment is over 30 days overdue. We may also close an account unused for 24 months with no paid balance on 30 days' email notice, or terminate for convenience on 60 days' notice.

12.4. Effect. On termination the Section 3 license ends, except open source licenses and, unless we terminate under Section 12.3(a), your license to the AgentD Core versions you have obtained, which continues under Section 3.1. For 30 days you may export User Content, read-only or, where that is unlawful or unsafe, by other secure means, even after a payment dispute or AI Provider demand; we then delete it under Section 6.6. Whoever terminates and why, unused purchased credits are refunded at the price paid, never forfeited. Unused Token Plan days are refunded pro rata if we terminate or end a paid service without your breach, or either party terminates under Section 5.3, 13.4, 13.6, or 13.7; otherwise there is no refund unless mandatory law requires one. Refunds, less only undisputed amounts you owe, go to your original payment method within 30 days. Mandatory law, including consumer, gift-card, and stored-value rules, prevails over this Section and the Billing Terms. Surviving: Sections 2; 6.1, 6.3, and 6.5 to 6.7; 6.2 until deletion under Section 6.6; 8 for prior use; 9 to 11; 12.4; 13; and 14.

13. Governing Law, Disputes, and General Terms

13.1. Governing Law. Delaware law governs this Agreement, without regard to conflict-of-law rules; the UN Convention on Contracts for the International Sale of Goods does not apply.

13.2. Disputes. Before suing, a party will send the other a dispute notice at its notice address (ours: Section 14) giving its name, contact details, the facts, and the relief sought, and the parties will try to resolve the dispute informally for 30 days after receipt, during which limitation periods are tolled to the extent law permits. This step does not apply to urgent interim or injunctive relief, or where delay would bar or prejudice a claim by law. Claims then go to the state or federal courts in Wilmington, Delaware, whose jurisdiction both parties accept; a Consumer may also sue in their local courts where local law allows.

13.3. Export Controls. You will comply with applicable export control and sanctions laws, including U.S. law, and represent that you are not in an embargoed country or on a restricted-party list.

13.4. Changes. We will notify you of every change by email or in the Services, and no change is retroactive. A material change (one that adversely affects your rights or obligations) needs 30 days' notice and your acceptance under Section 1.2; until then the prior version applies. Other changes take effect on the noticed date, at least 7 days later. If you do not accept a material change, the prior version continues for any prepaid period unless law or an AI Provider requirement prevents it, and we may then end your access with a refund under Section 12.4.

13.5. Entire Agreement. This Agreement, with the documents and obligations Sections 1.4 and 5.3 incorporate, is the entire agreement between us about the Software.

13.6. Assignment and General. Either party may assign this Agreement, with notice, to a successor to substantially all of its relevant business that assumes it in writing; other assignments need the other party's consent, not unreasonably withheld. The assignor is released only from obligations the assignee assumes. If we assign, you may terminate within 30 days of notice. An assignment or change of control does not change the terms that apply to Software versions you have already obtained. A failure to enforce is not a waiver. An unenforceable provision does not affect the rest.

13.7. Force Majeure. Neither party is liable for failures caused by events beyond its reasonable control, including AI Provider outages, if it gives prompt notice and reasonably mitigates, but payment for services received is not excused. If such an event makes the Services materially unavailable for over 30 consecutive days, either party may terminate the affected service.

14. Contact

Nebula Nova Inc., a Delaware corporation, Attn: Legal. Legal notices: contact@libraos.com. Security reports: contact@libraos.com. Support and review: contact@libraos.com.