Celeborn Code User Agreement

Version 2026-08-06 · source-available · celeborncode.ai

1. Celeborn Code is a tool

Celeborn Code has no mind, agency, or intent of its own. It is software that performs only the actions you turn on by installing it, wiring it into your coding agent, and running it. Every behavior described below is deterministic and configurable. By installing, wiring, and using Celeborn Code you instruct it to act on your behalf and direct it to perform the automations you leave enabled. Celeborn Code does not decide what to do — it carries out what you have configured.

The same is true of the AI model your coding agent runs: your model is yours, not ours. Celeborn Code is a harness around a model that you choose, you authorize, and you supervise. The model is not under Celeborn Code's control, and Celeborn Code assumes no responsibility for what it does.

2. What Celeborn Code does for you, and why

Each item below removes an approval click. All are enabled by default; you may opt out of any of them at install time or any time afterward by running celeborn consent.

  1. Pre-approve safe read-only commands and enable acceptEdits modeSo you stop clicking “Allow” for ls / grep / git-status and routine file edits.Safety: Your editor and the safe-listed commands then run without a per-action prompt. The allow-list is read-only/reversible commands only; acceptEdits lets the agent write files in the working tree without asking each time.
  2. Generalize your repeated approvals into reusable allow-rules (celeborn permissions)A command you approve more than once becomes a wildcard rule you never re-approve again.
  3. Capture and restore project context automatically across sessionsOrient load, per-turn capture, and checkpoint reminders — so a /clear or compaction never makes you re-explain the project.Safety: Project context is written to a local .context/ directory in your repo. Signing in uses your email for identity (GitHub can be linked optionally); your context stays on your machine unless you enable sync.
  4. Steer an un-approvable cd … > file write to the Write tool (PreToolUse guard)Turns a recurring manual approval into an invisible, statically-safe file write.
  5. Auto-allow a marked cd … > file write with no promptA command tagged # celeborn:allow-redirect runs without asking.Safety: Significant: a shell write whose target the permission system cannot statically verify then runs with no human check. This is OFF unless you keep it enabled; it exists for the rare case a shell redirect is genuinely required. The default posture denies the pattern and steers to the Write tool.
  6. Auto-launch your local kanban board on orientThe board comes up on a local port without a manual command.
  7. Auto-claim a task card when you paste its markerPasting a card assigns it to you without a separate claim step.
  8. Run tests / typecheck automatically after edits (opt-in via wire-quality)Surfaces failures without you remembering to run the suite.
  9. Sync features-only routing-outcome rows to power the Model MixerThe blended (lower) price tier is funded by this closed feedback loop; the routing discount is the tier price itself.Safety: Only features-only rows leave the machine — salted hashes and scalar features (class, model, rung, token/cost/outcome numbers), never your code, prompts, file paths, or card text. The salt is per-install, so rows are not cross-linkable between installs. Opting out flattens you to frontier-only routing defaults (you keep every capability; you lose the discount tier). Full detail: celeborncode.ai/model-mixer.

3. Safety considerations

Reducing prompts means Celeborn Code (and the coding agent it serves) takes some actions without asking you each time. You should understand and accept the following before enabling them:

4. Your control

You remain in control at all times. Run celeborn consent to review or change which automations are enabled. Manage the underlying agent permissions with your agent's permission settings (for example /permissions in Claude Code). Re-running celeborn wire never removes anything you added, and you can unwire or uninstall Celeborn Code to stop all of its behavior.

Control and responsibility travel together. Because all authority over your coding agent and its model flows from you and only you, responsibility for how that authority is used rests with you and only you — none of it rests with the coding harness.

5. Acceptance

By typing your name into the consent prompt you confirm that you have read this agreement — including the Terms of Use below — that you understand the automations and safety considerations above, that you accept sole responsibility for supervising your coding agent and its model, and that you direct Celeborn Code to perform the automations you have left enabled. Your name, the time, the agreement version, and your opt-out selections are recorded locally to ~/.context/consent.json.

6. Terms of Use — parties and definitions

The sections from here to the end of this page are the Terms of Use for Celeborn Code. They are part of this User Agreement: by installing, wiring, running, or otherwise using Celeborn Code, or by accepting the consent prompt, you agree to them. If you do not agree, do not use Celeborn Code.

7. Your Model is under your authority, not ours

Celeborn Code is a coding harness. It does not create, host, train, operate, or control Your Model, and it cannot. You choose Your Model, you connect it, you grant every permission it holds, and you supervise everything it does. You are the sole authority-grant holder for Your Model and for every action taken through it, and absolutely zero responsibility for Your Model's behavior rests with the coding harness, with the Providers, or with any Associated Person.

You agree that you are responsible for supervising Your Model at all times while using Celeborn Code; that Your Model is not under the control of Celeborn Code; and that anything Your Model reads, writes, deletes, executes, transmits, or purchases — whether you were watching or not, whether you intended it or not — is attributable to you as the person who granted it authority, and to no one else.

8. License and acceptable use

These Terms of Use supplement, and do not replace, the project's source-available license (BUSL 1.1); where they conflict, the license governs your rights in the source code. You agree to use Celeborn Code only in compliance with applicable law, only on systems and repositories you are authorized to use, and not to use it to infringe the rights of others. You — not the Providers, not any Associated Person, and not the coding harness — are accountable for what you and Your Model do with it.

9. Easy and Hosted Plan subscriptions

Celeborn Code's local engine — memory, context capture, the board, and multi-agent coordination — runs on your machine. Easy grants access to that local product for $20 per month. You supply and pay for a compatible Claude Code or Codex account, and that provider's terms govern its model service. Easy includes no model usage from Celeborn, and Celeborn does not receive or broker your provider login credentials.

Junior, Senior, and Chief are Hosted Plans. They add a curated Trusted Flow model mix that runs through our AI Gateway, and each carries a monthly usage budget. The budget terms below apply only to a Hosted Plan.

10. Data processing and the model providers

When you run a paid Trusted Flow mix, it sends your prompts and the code context each step needs to the model providers that run that step. We name those providers so you know who processes that content before you decide to route work to them.

Calls reach these providers through our AI Gateway, which routes to them on your behalf. Your local engine still keeps your code, your memory, and your .context/ files on your machine; it is the paid mix, when you run it, that sends prompt content to the providers above. If routing work to a PRC-based provider does not fit your organization, do not subscribe to the paid Trusted Flow mix — your local engine and your own directly-connected coding agent do not route to it.

11. “As Is” — no warranty

CELEBORN CODE IS PROVIDED STRICTLY “AS IS” AND “AS AVAILABLE”, WITH ALL FAULTS AND WITHOUT WARRANTY OF ANY KIND. TO THE MAXIMUM EXTENT PERMITTED BY LAW, THOT TECHNOLOGIES LLC, CLOUD DANCER, AND ALL ASSOCIATED PERSONS DISCLAIM ALL WARRANTIES, EXPRESS, IMPLIED, STATUTORY, OR OTHERWISE — INCLUDING, WITHOUT LIMITATION, ANY IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE, NON-INFRINGEMENT, ACCURACY, QUIET ENJOYMENT, AND ANY WARRANTY ARISING FROM COURSE OF DEALING OR USAGE OF TRADE. NO ONE WARRANTS THAT CELEBORN CODE WILL BE UNINTERRUPTED, ERROR-FREE, OR SECURE, OR THAT YOUR MODEL WILL BEHAVE AS YOU EXPECT.

No advice or information, whether oral or written, obtained from the Providers, any Associated Person, Celeborn Code itself, or Your Model creates any warranty not expressly stated here. In particular, nothing about the harness's automations, permission baselines, or safety features is a promise that Your Model will act safely or correctly — supervising Your Model remains yours to do, and yours alone.

12. Zero liability

YOU AGREE THAT NEITHER THOT TECHNOLOGIES LLC NOR CLOUD DANCER HAS ANY LIABILITY WHATSOEVER FOR ANYTHING THAT HAPPENS WHILE YOU ARE USING CELEBORN CODE. TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE PROVIDERS AND ALL ASSOCIATED PERSONS SHALL NOT BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, EXEMPLARY, OR PUNITIVE DAMAGES, NOR FOR ANY LOSS OF DATA, CODE, PROFITS, REVENUE, GOODWILL, OR BUSINESS, NOR FOR ANY COST OF SUBSTITUTE GOODS OR SERVICES, ARISING OUT OF OR RELATING TO CELEBORN CODE OR THESE TERMS — WHETHER IN CONTRACT, TORT (INCLUDING NEGLIGENCE), STRICT LIABILITY, OR ANY OTHER THEORY, AND EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

THIS INCLUDES, WITHOUT LIMITATION, ANYTHING DONE OR NOT DONE BY YOUR MODEL. THE MODEL YOU RUN IS NOT UNDER THE CONTROL OF CELEBORN CODE; YOU GRANTED ITS AUTHORITY AND YOU BEAR ITS CONSEQUENCES. NO RESPONSIBILITY OF ANY KIND ATTACHES TO THE CODING HARNESS. IF, NOTWITHSTANDING THE ABOVE, ANY LIABILITY IS FOUND, THE AGGREGATE LIABILITY OF THE PROVIDERS AND ALL ASSOCIATED PERSONS TOGETHER SHALL NOT EXCEED THE AMOUNT YOU PAID FOR CELEBORN CODE IN THE THREE MONTHS BEFORE THE CLAIM AROSE, OR TEN U.S. DOLLARS ($10), WHICHEVER IS GREATER. Some jurisdictions do not allow certain exclusions or limits; in those jurisdictions these apply to the fullest extent the law permits.

13. Indemnification — you hold us harmless

You agree to fully indemnify, defend, and hold harmless each of the following, separately and independently, from and against any and all claims, demands, actions, suits, losses, liabilities, damages, judgments, settlements, fines, penalties, costs, and expenses (including reasonable attorneys' fees) arising out of or relating to your use of Celeborn Code, the conduct of Your Model, your violation of these Terms, or your violation of any law or the rights of any third party:

Each of these three indemnities is independent: the failure or unenforceability of one does not diminish the others, and each indemnified party may enforce its indemnity directly. These indemnities cover, without limitation, anything Your Model does under the authority you granted it — because you are its sole supervisor and sole authority-grant holder, claims arising from its conduct are yours to answer for, not the coding harness's.

14. Assumption of risk

Working with AI coding agents means giving software the ability to read, write, and execute on your machine. You knowingly and voluntarily assume all risk of doing so. You acknowledge once more — in plain words this time — that the deal is: your model, your permissions, your supervision, your responsibility. Celeborn Code hands you levers; it does not hold them for you, and it answers for none of what happens when you or Your Model pull them.

15. General terms