Entire launches a GitHub rival where agents score the change and humans stop reading every line

The premise holds up. Entire is real and in preview, but it is not alone: Cursor Origin and a GitLab beta are already competing, and GitHub is the incumbent. "Agent-native code hosting" is a contested category among well-funded players, not an open one. The security-gate idea is the weaker of the two. Snyk, Semgrep and Endor Labs all ship checks that run inside the agent loop through MCP servers and IDE plugins. The Reva paper itself only reaches about 60% accuracy. A new entrant would be a better model competing against established distribution. The one opening is making the gate mandatory and enforced by the platform, rather than something the agent may or may not call. A forge like Entire or Origin could add that cheaply itself. The audit-trail idea is less contested. Provenance capture exists, but only as early open-source or early-access tools: Git AI, agentdiff, Lineage Lens, aigit and Bitloops. Entire's own CLI already logs agent sessions. What I could not find is a product that turns that data into evidence auditors accept, mapped to SOC 2 CC8.1 and to SOX segregation-of-duties and change-management controls. Such a product would record which scoring agent approved a change, its model version and evidence, and which lines no human read, and would come with attestation reports a bank's auditors and regulators will sign off on. A neutral layer across GitHub, Entire, Origin and GitLab, sold to bank and payments compliance and engineering leadership, could win. Two risks stand out. First, the forges own the underlying data and may bundle a basic version. Second, I found no regulator or AICPA guidance on code merged without a human reading it, so the startup would have to shape what auditors accept, working closely with audit firms, rather than rely on an existing standard.

• Entire (Thomas Dohmke)

  The company in the premise. It is real. It launched in February 2026 with a $60M seed and is in preview as a distributed Git network with regional mirrors built for agent traffic. Its CLI records agent sessions and decisions next to commits, so it already captures the raw provenance that an audit trail would be built on.

• Cursor Origin

  An agent-native Git forge that went into early beta for paid Cursor plans on August 17, 2026. It has repos, pull requests, two-way GitHub sync and the Graphite review tool Cursor acquired. It is the most direct rival to the hosting-plus-review idea.

• GitLab agent-oriented source control

  Reported to have been in private beta since June 2026. Incumbents are building their own agent-era forges.

• Git AI (open source)

  A Git extension that stores which lines are AI-written as Git Notes on each commit, with an enterprise dashboard in early access. Reviewers describe it as a provenance layer, not a compliance product.

• agentdiff

  A Git-native provenance tool. It records the agent, model, prompt excerpt and exact line ranges, and signs each record with ed25519. That makes it close to tamper-evident evidence, but it has no auditor-facing layer.

• Lineage Lens

  Open-source tool launched on Product Hunt. It captures provenance through a proxy and claims to flag AI-generated code in sensitive paths that no human reviewed. This is the closest match to the unread-lines idea, but it is early and the claims are unverified.

• Bitloops

  A vendor pitching audit trails for AI-assisted development (prompt, model version, constraints and reviewer for each change). It is positioned at governance buyers.

• Reva (research paper)

  The source of the Reva-style idea: an academic detector that checks each function as a coding agent writes it, at about 1 second per function and 60% accuracy on PrimeVul. It is research, not a product, so turning it into a product is open, but the accuracy is modest.

• Snyk MCP / Snyk Code

  A local MCP server that agents call to scan code they just wrote. It covers code, dependencies, containers and infrastructure-as-code. It is the incumbent default for checks inside the agent loop.

• Semgrep MCP

  The official MCP server lets an agent scan and fix generated code before it commits. It is also embedded in Replit's Security Center for real-time feedback. A known weakness is that the agent only scans when it decides to.

• Endor Labs AURI

  Launched March 2026 as a plugin, MCP server and CLI that flags reachable vulnerabilities inline in Cursor and Claude Code.

• Reware Labs

  A startup selling agentic code security that steers coding agents toward safer code as they write it.

Checked 2026-10-27