Agent infrastructure, built in the open
rossoctl is an open source research project building the infrastructure AI coding agents need to run for real: a dense runtime where sessions resume exactly where they left off, a proxy that shows what your coding agent sends and what it costs, an event-driven wake path, automated benchmarking, and state management for distributed agents. Around that core we work on context, access control, and data governance, and we run the org itself on the automation we build.
Core projects​
The projects we are actively building and shipping.
Moca
A dense, warm process-manager runtime for stateful AI coding agents — sessions resume exactly where they left off. Agent state is durable, tool execution runs in isolated sandboxes, and one supervisor serves many sessions per warm worker.
Cortex
See what your coding agent actually sends — and pay less for it. A proxy in the agent's request path that shows every model call, tool call, and cost as it happens, and strips the tool definitions the agent never calls. Cortex Local runs on individual, end-user machines; Cortex Remote, deployed along the harness, is being explored.
Eventing
The signed-event agent wake path: a Kafka message wakes a coding agent, and every step of its run streams back out as CloudEvents — with identity and signing on the event path.
AutoBench
Automated benchmarking of agentic AI workloads: deploys a benchmark workload — a tool-calling agent paired with the benchmark's MCP server — runs it, collects per-task telemetry, and publishes the results as durable artifacts.
State management
Where agent state lives, who owns it, and what happens when a process dies mid-task — demo flows exploring durability and passability for distributed agents.
Context & governance​
Context, access control, and data governance for agent workloads.
Context Guru
Provider-agnostic context engineering for LLM agents — lower token cost and latency without sacrificing accuracy. Ships as plugins for Claude Code and Codex.
AIAC
AI-based access control: translates natural-language policy into enforced permissions on OPA, driven by Keycloak identities.
Data governance
Data-aware governance and security: observes how data flows across agents, tools, and sessions, and derives interactions, payload classifications, and data lineage from the raw stream.
Context Service
Agent context infrastructure: durable storage and lifecycle management for agent state, memory, knowledge, workspaces, and artifacts — for local harnesses and container-native platforms alike.
Automation & repo maintenance​
The automation that keeps the org and its repositories healthy.
Automation
Automated maintenance programs for the rossoctl org, built on a scanner/fixer pattern.
RepoMan
Automated maintenance programs — PR review, link health, dependency bumps, weekly reports — across GitHub repos you enroll once. Install once, enable programs on the fly.