Execution infrastructure for coding agents
NullDeref gives your coding agents a working replica of each repository’s real target. They build, debug, test, and benchmark there autonomously—then return reviewable work.
Runs inside your controls PRs remain the review gate No per-seat tax
A working environment for every target
Agent compatible
Claude Code, Codex, and Cursor already know how to write code. NullDeref supplies the part they were missing: a dependable environment where the code can compile, boot, fail, recover, and prove the result.
Autonomous execution
The agent changes the code, compiles it, diagnoses the break, runs the test suite, and measures the result. Winners become pull requests. Regressions are reverted. Engineers review evidence instead of supervising terminal sessions.
The platform
Start with one repository. Add the operating layer your team needs as agent work scales across the organization.
Autonomous execution
Loop
Turn an engineering goal into a measured pull request. The agent builds, debugs, tests, and benchmarks until the acceptance gate passes.
Explore Loop →Agent operations
Sync
Automatically roll proven AI coding techniques, harnesses, and agent patterns into every compatible repository—keeping every team current before competitors gain an adoption lead.
Explore Sync →Organization readiness
Scope
See which repositories are agent-ready, where execution still depends on tribal knowledge, and what each agent can reach at runtime.
Explore Scope →Token economics
Meter
Measure tokens per successful task, find repeated discovery work, and turn it into reusable repository knowledge.
Explore Meter →Autonomous red teaming
Red
Boot an isolated production-shaped simulation. Frontier security harnesses probe the running target, prove exploit impact, and return replayable evidence.
Explore Red →Engineering intelligence
Pulse
Ask what a branch is for, where a feature is blocked, or whether a POC worked. Get a current answer grounded in repositories, tickets, CI, and agent runs.
Explore Pulse →Measured result
The agent was given a straightforward goal: speed up an image-processing task without changing its output. It ran the code, measured each attempt, kept improvements, and discarded regressions until the performance gate passed.
Pricing
Copilot, Cursor, and Claude charge the human. NullDeref charges the repository—the durable unit that needs an environment, policy, and execution history. Every engineer on a connected repo is included.
Self-serve
Repo
$49 / repo / mo
A maintained execution environment for one GitHub or GitLab repository. Everyone on that repo is included.
Volume
Org
Contact
Twenty-five or more repos. One picture of the org, not a stack of invoices.
Custom
Enterprise
Contact
SSO, Jira, isolated Red labs, and terms that survive a security review. The floor is a conversation.
Published list price, not a negotiated floor. Agent usage from Anthropic, OpenAI, or Cursor is billed by those vendors. Enterprise controls, private networking, and custom terms are scoped separately.
Start with one repository
Tell us where you want to start. We’ll confirm the target environment, repository controls, and the first acceptance gate with your team.
Use a work email so we can discuss the repository and its target environment.