Macroscope

Connections cover GitHub, Slack, Linear, Jira, Sentry, PostHog, LaunchDarkly, BigQuery, Google Cloud Logging and Amplitude, and the agent can create Linear issues and sub issues as well as read them. Workspace admins can add any MCP server with a URL and OAuth credentials or headers, review the tools it exposes, switch off any the agent should not use and limit a connector to named users. On GitHub the agent creates branches, writes code, opens pull requests, updates existing branches and adds reviewers.

Sourcedocs.macroscope.com/integrationsread 2026-11-08

Check Run Agents can wait for or require one another, with cycles caught, and a per code object mode runs a separate agent for each changed function or class, up to 20 at once. The review itself gathers context in several agentic steps, and the Ultra mode runs more agents in parallel.

Fix It For Me tries again with further commits when GitHub Actions fail, the CLI autoloop fixes and rechecks for up to five rounds, and Macros pair a trigger with plain English instructions and a delivery target. Murmur adds flights, Markdown graphs of agent tasks with gates for CI, review and merge, behind its waitlist.

Sourcedocs.macroscope.com/check-run-agents/configurationread 2026-11-08

The agent searches code across every connected repository along with git history, pull request discussion and reviews, connected tools such as issue trackers, documentation and the web. The correctness review explores the code around each change before it comments, and Check Run Agents start already holding the pull request's discussion and can pull in named files from the repository. Status answers questions about what changed across repositories with code and issue tracker context, guided by a product overview the team writes.

Sourcedocs.macroscope.com/agent.mdread 2026-11-08

Approvability starts switched off and can run in a preview mode that shows what it would approve without approving anything. Teams set strict or permissive mode, the lowest severity that blocks approval, and paths it must never approve, always approve or only approve, and any change to its own rules file always waits for a person. Reopening a blocking thread withdraws an approval. Fix pull requests aimed at main always need a manual merge, Check Run Agents block only when set to fail on a required check, and the agent's calls to outside tools are read only by default.

Sourcedocs.macroscope.com/approvability.mdread 2026-11-08

Sign in runs through GitHub, including Enterprise Managed Users through the organization's identity provider, with GitHub organization admins and delegated admins running the workspace, repository write access required for review features and API keys limited to admins.

SOC 2 Type II reports are shared through a request access flow on the trust center, along with penetration test results and a subprocessor list. Customer source code is not used to train models, and content is deleted within 30 days of termination. There is no SCIM, and the terms allow training on deidentified content other than source code.

Sourcemacroscope.com/content/ai-code-review-security-soc-2-data-privacyread 2026-11-08

The Approvability dashboard lists every pull request it judged with the verdict and the agent's reasoning, and a review activity log records feedback with CSV export. Each Check Run Agent shows its tool calls and billed cost in the GitHub checks tab. Hourly or daily S3 exports deliver check run logs with tool calls, review analytics and user activity records naming the actor, source IP and action, and usage charts break spend down by day and by agent.

Sourcedocs.macroscope.com/s3-exports/formatread 2026-11-08

Within a pull request, the review keeps the code research it gathered and tracks which findings were resolved, linking each one to the commit that fixed it. No learned preference or feedback carries from one pull request to the next, and standing guidance lives in rules files and custom instructions that a team edits by hand. Murmur can resume a stopped agent on the same branch with its full conversation history, though only for teams admitted from its waitlist, and no scope, lifetime or clearing control is described for any of it.

Sourcedocs.macroscope.com/correctnessread 2026-11-08

The service runs on Google Cloud as its main host with AWS and Microsoft Azure, and customers can keep model data and inference in the US for some providers. Code review is cloud hosted only, with no self hosted, private cloud or air gapped option, and it supports GitHub alone, with GitLab on the roadmap. Murmur agent machines can run in a customer's own Google Cloud or AWS account, set up with Terraform, but Murmur is still in early preview behind a waitlist.

Sourcemacroscope.com/content/ai-code-review-enterprise-deployment-complianceread 2026-11-08

Check Run Agents and Macros can both be created from prebuilt templates. Ready made Macros include daily stand up notes, weekly product updates, incident alerts to PagerDuty, linking pull requests to tickets, alerts on key pull requests and release branch diff summaries, each a working agent with a trigger, instructions and a delivery target. The correctness check and the default Approvability criteria work out of the box, and Murmur adds a verifier agent guide and agent personas for its waitlisted users.

Sourcedocs.macroscope.com/macros.mdread 2026-11-08

Reviews start when a pull request opens, receives a push or leaves draft, and Check Run Agents also fire on labels added or removed, mentions of the Macroscope app and check reruns. The agent answers Slack messages and mentions, GitHub comments and a webhook endpoint secured with a shared secret, and Macros run daily, weekly or every two weeks or on each new commit or pull request. A GitHub Action can call it from any workflow, and Slack subscriptions post commits, pull requests and build failures filtered by path.

Sourcedocs.macroscope.com/check-run-agents/triggersread 2026-11-08

Each Check Run Agent names its own model, from Claude Opus 5.5 by default to other Claude Opus and Sonnet versions, OpenAI GPT models through GPT 6.1 Terra, Grok 4.6, Meta's Muse and open models such as Kimi K3, GLM 5.3, DeepSeek and Qwen hosted on Fireworks, with per token prices listed. The correctness review picks its models by the detection mode a team sets, from Budget to Ultra. Custom endpoints and self hosted models are not supported, and the bring your own key option in Murmur sits behind its waitlist.

Sourcedocs.macroscope.com/model-pricing.mdread 2026-11-08

A REST API signed with workspace keys returns workspace, review, commit, contributor, billing and usage data, and every endpoint is read only, so an outside program can report on the agent's work but cannot start a review, change a setting or create anything through it. A CLI brings review and an autoloop fix command into Claude Code, Codex, Cursor and OpenCode. The core product ships no SDK and no MCP server of its own, and the MCP server with 22 tools and the separate API belong to Murmur, which is open only to waitlisted teams.

Sourcedocs.macroscope.com/apiread 2026-11-08

Approvability's preview mode runs on real pull requests and shows what it would have approved, so a team tests its approval policy before granting live authority. Rules in the .macroscope folder are read from the latest commit on a pull request, so a new check or instruction can be tried there before it merges, and the CLI runs the same review locally before a push. Review statistics with a ten week feedback trend and a dashboard of verdicts with reasoning show how the agent performs over time.

Sourcedocs.macroscope.com/approvability.mdread 2026-11-08

The review and the agent search the web and fetch pages but never drive a browser. Murmur runs each coding agent on its own cloud machine that can start the full stack, interact with the app's interface, inspect logs and attach screenshots or recordings as proof in its pull requests, with Playwright installable on the machine. Murmur is in early preview and open only to teams admitted from a waitlist, so this capability is not on general sale.

Sourcemacroscope.com/content/murmur-faqread 2026-11-08