Skip to content

From plan to pull request.

Octavo plans and edits in the editor and the terminal, runs tasks in cloud VMs and reviews every pull request your team opens.

Request a demo

Agent shows every step. It plans, edits across files, runs commands and tests, fixes what fails and hands you a diff to review.

  • Plan mode asks first.

    Plan mode writes a plan with to-dos and asks clarifying questions. It builds only when you say so.

  • Rewind any step.

    Every agent step has a checkpoint. Rewind to one and your files return to that point.

  • Runs your tests, fixes failures.

    The agent runs your tests, reads each failure and fixes it, then shows the whole change as one diff.

  • Checks the app it changed.

    The agent opens your running app in the built-in browser, clicks through it and takes screenshots to verify UI changes.

Tab predicts your next edit. Octavo Tab, an in-house model built for low latency, proposes multi-line edits as you type. Press Tab to accept.

How the Tab model works
  • Suggests several lines at once.

    Tab proposes a whole function body, a block of related lines or a rewrite of the line you are on.

  • Jumps to the next edit.

    After you accept an edit, Tab marks the next line that needs a change. Press Tab again to jump there.

  • Adds the imports.

    When a suggestion uses a symbol from another file, Tab adds the import at the top of the file too.

  • Follows your recent edits.

    Tab reads the changes you just made and proposes the same change in the other places that need it.

The same agent in your terminal. The Octavo CLI plans, edits files and runs your tests, and prints each change as a diff.

Platforms and requirements
  • A session in any shell.

    Start a session in any repository and steer the agent as it works, with the same models as the editor.

  • Headless in scripts and CI.

    With octavo -p, the CLI runs one task without a session and exits. The result comes back as text or JSON.

  • Over SSH and in containers.

    Run it on a build server or in a container image your team approved, and the agent uses that machine’s files and tools.

Each task gets its own VM. A cloud agent works with your repo, dependencies and tools, and returns a branch and a pull request.

Run cloud agents in your own VPC
  • Many agents at once.

    Run many cloud agents side by side, across repositories. Their processes and test runs stay out of each other’s way.

  • Start from Slack, GitHub, Linear or the web.

    Mention @Octavo in Slack, comment on a GitHub issue or pull request, start from a Linear issue or type the task in Octavo Cloud.

  • Pull requests that show the work.

    Each pull request has a summary, test results and screenshots or a short screen recording of the change.

  • Move work between laptop and cloud.

    Send a task from the editor or CLI to the cloud, and bring it back to finish locally.

A reviewer on every pull request. Octavo Review reads pull requests on GitHub and GitLab and comments inline on logic bugs, security issues and regressions.

How Review decides what to flag
  • Each comment comes with a fix.

    Apply the suggested change, or choose Fix in Octavo and an agent makes the fix.

  • Learns your team’s rules.

    Octavo Review checks each change against your project and team rules. It is tuned to post few comments, each one worth reading.

  • Make it a required check.

    Add Octavo Review as a required check. When it finds nothing to fix, it stays quiet and the check passes.

Pick the model for the task. Choose per request from frontier and open-weight models, or from Octavo’s own Verso 1.5.

See included usage
Models you can use in Octavo
ProviderModelGood for
AnthropicOpus 5.5Large features and refactors across many files
Fable 5.1MaxHard problems that need long reasoning
Sonnet 5.5FastEveryday coding with quick turnaround
OpenAIGPT-6.1 SolEveryday agent tasks across a codebase
GPT-6 AstraMaxLong tasks over very large contexts
GPT-6 LunaFastQuick questions and small edits
GoogleGemini 3.1 ProPlanning and reasoning over long inputs
Gemini 3.8 FlashFastFast work on long, multi-step tasks
SpaceXAIGrok 4.7General coding and broad-knowledge questions
MetaMuse Spark 1.3General coding and explanations
Moonshot AIKimi K3Long-context work on an open-weight model
DeepSeekDeepSeek-V4-ProGeneral coding on an open-weight model
Z.aiGLM-5.3Agentic coding on an open-weight model
OctavoVerso 1.5FastQuick, accurate multi-file edits
Octavo TabFastNext-edit predictions in Tab
  • Leave it on Auto.

    Auto, the default, picks a model for each request. Choose a specific model in the picker at any time.

  • Verso 1.5, made by Octavo.

    Octavo’s fast agentic coding model, trained for quick, accurate multi-file edits. On Auto, it is the default for speed-sensitive work.

    Introducing Verso
  • Bring your own keys.

    Admins can add your organization’s own API keys for OpenAI, Anthropic, Google, SpaceXAI, Azure OpenAI, AWS Bedrock or Vertex AI.

Context you control. A semantic index, your rules, Memories and MCP servers give the agent your code, conventions and tools.

What the index stores
  • Indexes the whole repository.

    The agent and search find code by meaning, in monorepos too. The index respects .gitignore and .octavoignore.

  • Rules the agent follows.

    Write project rules as Markdown in .octavo/rules. Team rules sync from the dashboard to everyone on the team.

  • Memories you approve.

    The agent proposes a Memory when it learns something about your project. Nothing is saved until you approve it.

  • Connect tools with MCP.

    Connect databases, browsers, docs and issue trackers through Model Context Protocol, and the agent uses them as it works.

How teams use Octavo.

  • “Our monorepo has dozens of services and three naming schemes. The index finds code by what it does, so the agent starts in the right service. Our rules file says which naming scheme wins and where migrations go, and the agent’s diffs now follow both without a reminder.”

    Marta HauglandStaff Engineer · Tarnwick Logistics
  • “Octavo Review reads every pull request in our Swift and Kotlin repositories. It caught a change that reset a user’s streak when their phone switched time zones, and the comment came with the fix. Our team rules tell it to skip formatting, so its comments are about behavior.”

    Linh PhamMobile Lead · Kettlebrook
  • “Our nightly pipeline checks run in CI. When one fails, the job runs octavo -p on the log and writes a summary that names the stage that broke and why. Most mornings I start from that summary instead of scrolling through thousands of log lines.”

    Priya RaghunathanML Engineer · Driftlake

See it on your own codebase.