# ADR 045: Qodo

- HTML version: https://robbiepalmer.me/projects/personal-site/adrs/045-qodo
- Project: Personal Site (https://robbiepalmer.me/projects/personal-site.md)
- Status: Accepted
- Date: 2026-06-21

# Context

[ADR 009: CodeRabbit](/projects/personal-site/adrs/009-code-rabbit) records the decision to use automated AI review on Pull Requests. [ADR 041: Gemini Code Assist](/projects/personal-site/adrs/041-gemini-code-assist), [ADR 042: Greptile](/projects/personal-site/adrs/042-greptile), and [ADR 044: Codex Code Review](/projects/personal-site/adrs/044-codex-code-review) record why multiple review providers improve coverage and resilience. This ADR does not repeat that rationale.

The available review capacity is no longer sufficient for the project's development pace:

1. **CodeRabbit** now assigns this repository the bottom of its variable open-source allowance: one review per hour. Each reviewed push consumes that allowance, so a Pull Request developed through several commits can be delayed repeatedly rather than receiving one allowance for the Pull Request as a whole.
2. **Gemini Code Assist** consumer reviews will cease on July 17, 2026.
3. **Codex code review** has a finite allowance that is being exhausted.
4. **Greptile** remains useful as an additional opinion but does not provide sufficient coverage or feedback quality to sustain the workflow by itself.

The constraint is therefore both capacity and review quality. Replacing all existing services with the remaining free reviewer would reduce confidence and slow delivery.

# Decision

I will add **[Qodo](https://www.qodo.ai/)** as another automatic Pull Request reviewer.

Qodo's **[open-source programme](https://www.qodo.ai/solutions/open-source/)** provides its GitHub review integration free for public open-source repositories. Its multi-agent workflow prioritises findings by severity, validates changes against requirements, and can produce implementation-ready prompts for coding agents. This gives Qodo a different review emphasis from CodeRabbit's broad walkthrough and accumulated learnings, Codex's high-severity default, and Greptile's graph-based context.

Qodo's open-source service is powered by Gemini models through Google Cloud. It therefore restores review capacity after the Gemini Code Assist consumer integration closes, but does not restore the same degree of foundation-model diversity. The expected value comes from Qodo's review agents, context, prompts, prioritisation, and workflow rather than assuming that it uses a distinct model family.

This is an accepted evaluation rather than a claim that Qodo can replace CodeRabbit. Its practical value will be judged from the relevance, novelty, and accuracy of its comments on this repository.

# Alternatives

## Pay for CodeRabbit

* **Pros**: Retains the reviewer that currently provides the strongest coverage and preserves its accumulated learnings. **[CodeRabbit Pro](https://docs.coderabbit.ai/management/plans)** costs $24 per developer per month when billed annually or $30 month-to-month and allows five Pull Request reviews per rolling hour. Pro+ costs $48 per month annually or $60 month-to-month and allows ten reviews per hour. Both allow 300 files per review, compared with 150 on the open-source tier.
* **Cons**: Adds a subscription to restore capacity that was previously available to this open-source project. Five reviews per hour may still throttle periods with more than one reviewed push every twelve minutes. Pro+ doubles the price primarily for ten reviews per hour and additional planning, test-generation, conflict-resolution, and other workflow features that are not required for this decision. Either plan can exceed its hourly allowance through a usage-based add-on, but overflow costs $0.25 per reviewed file and can make large reviews expensive. The tighter open-source limit and increasing emphasis on paid planning and enterprise administration features also reduce confidence that its future product direction will remain aligned with a fast-moving solo open-source workflow.
* **Decision**: Deferred, not rejected. Pro is the appropriate first paid tier if Qodo and the remaining free reviewers cannot sustain review quality: it includes all ongoing review functionality and raises capacity to five reviews per hour. Pro+ is justified only if the workflow regularly exceeds that limit or its additional workflow features become valuable; otherwise occasional overflow credits are a more targeted option.

## Buy Codex Credits or Upgrade ChatGPT

* **Pros**: Extends an already configured reviewer without adding another GitHub App or review format. **[ChatGPT Plus and Pro users can buy credits](https://help.openai.com/en/articles/12642688-using-credits-for-flexible-usage-in-chatgpt-freegopluspro-sora)** after exhausting included Codex usage, so review capacity can be extended without changing subscription. Alternatively, ChatGPT Pro costs $100 per month for five times the standard Plus usage or $200 per month for twenty times the usage.
* **Cons**: OpenAI does not publish a fixed number of GitHub reviews for Plus or either Pro tier, so the plan multipliers do not establish a guaranteed review count. **[Code review uses GPT-5.3-Codex and token-based credits](https://help.openai.com/en/articles/20001106-codex-rate-card)**, making the cost depend on the Pull Request's input, cached input, and output tokens. Purchased credits are shared with other supported Codex and ChatGPT features rather than reserved for review, and their cash purchase options are shown in the account rather than published as a stable price per credit. Codex also has a narrower high-severity focus and has not matched CodeRabbit's breadth on its own.
* **Decision**: Rejected as the primary response. Buying credits on Plus is the most targeted Codex overflow option, but its variable cost and narrower reviews make it less predictable than adding a free reviewer. Upgrading solely for review would add $80 or $180 per month over Plus without a documented review count; if a paid reviewer becomes necessary, CodeRabbit Pro is cheaper and has explicit hourly limits.

## Greptile Only

* **Pros**: Avoids adding another provider and retains whole-codebase graph analysis.
* **Cons**: Its reviews have not been useful or comprehensive enough to sustain the project's delivery pace alone, and observed overlap with Gemini Code Assist reduces the value of relying on it as the only remaining reviewer.
* **Decision**: Rejected. Greptile remains complementary rather than sufficient on its own.

## Reduce Review Frequency

* **Pros**: Fits development within the remaining allowances without another integration or subscription.
* **Cons**: Either batches unrelated changes to conserve reviews or permits commits to proceed without timely feedback. Both weaken the review workflow and slow iteration.
* **Decision**: Rejected. Reviewer constraints should not dictate commit boundaries or reduce review coverage.

# Consequences

## Positive

* **Restored Review Capacity**: Another automatic reviewer reduces delays caused by CodeRabbit and Codex limits and the Gemini Code Assist sunset.
* **No Additional Cost**: Qodo is free for eligible public open-source repositories.
* **Different Review Emphasis**: Severity prioritisation, requirement validation, and agent-ready remediation prompts provide another angle on each change.
* **Provider Resilience**: Review coverage no longer depends on any one remaining service increasing its allowance or maintaining its current terms.
* **Evidence Before Purchase**: Qodo can be evaluated before deciding whether CodeRabbit's stronger paid service is worth its subscription cost.

## Negative

* **Model Correlation**: Qodo's open-source programme uses Gemini models, so its findings may overlap with Gemini Code Assist or other Gemini-backed reviewers.
* **Additional Noise**: Another automated reviewer can duplicate, contradict, or add lower-value comments that require human judgement.
* **Unproven Repository Value**: Qodo's advertised review quality and different emphasis must be validated through actual Pull Requests.
* **Programme Dependency**: The no-cost decision depends on Qodo continuing its free public open-source programme and this project remaining eligible.
* **Repository Access**: Qodo must be granted sufficient GitHub access to read the repository and post Pull Request feedback.
* **More Configuration**: Another reviewer introduces its own settings, commands, feedback conventions, and operational limits.

---

Markdown index of this site: https://robbiepalmer.me/llms.txt