Protocore
A protocol-first runtime for stateful, tool-using AI agents: strict core contracts, durable run state, observability and real-time SSE streaming. It deploys entirely on your own infrastructure and ships with a reproducible-research publication site. Flagship project — founder / solo.
SOURCE: MANUAL README
Protocore is a self-hosted AI-agent orchestration platform for organizations that cannot hand data to external SaaS: banks, the public sector, education, and industry. It deploys entirely on the customer's infrastructure. Public page: protocore.ascorblack.ru.
Role: founder / solo developer — I designed and built the core runtime, enterprise backend, sandbox, chat and admin frontend applications, and the Helm / Kubernetes deployment.
Architecture
Protocol-first: the core defines ~15 Protocol interfaces (ILLMProvider, IRunStore, IToolRegistry, IMemory, IWorkspace…), and concrete adapters live outside the core. The direction of dependencies is strictly enforced by guard tests: the core imports nothing from the enterprise, frontend, or deploy layers.
- protocore — a clean core in Python 3.12+: contracts + a ReAct runtime, an immutable orchestration loop with budget guardrails, subagent orchestration (leader / parallel / auto-select), context compression. Open open-core version: github.com/ascorblack/protocore.
- protocore-enterprise — the reference backend: FastAPI with SSE streaming, RBAC with 6 scoped roles, access plans with quotas and rate limits, LLM routing with health monitoring and a circuit breaker, a versioned control plane in Postgres.
- protocore-sandbox — isolated execution of shell/Python in per-session Kubernetes pods under gVisor.
- protocore-chat / protocore-dashboard — an end-user chat (SSE streaming, session branching) and an admin console on Next.js 15 / React 19.
- protocore-autonomous — scheduled autonomous tasks and multi-step workflows with leader election.
- Dynamic tooling subsystem — a tool is a single
.pyfile; CI publishes an immutable revision, and activation is per-scope with no backend redeploy.
Self-hosted / local LLMs are supported: vLLM, Ollama, and OpenAI-compatible providers.
Maturity
- ~15 repositories, ~440k lines of Python + TypeScript;
- 7,500+ tests across the core and enterprise packages; core coverage 96.01%, enterprise 84.07%;
- ruff / mypy / bandit / pip-audit — clean; live-backend integration tests with Docker;
- fully self-owned infrastructure: GitLab, Harbor, k3s, Helm, CI/CD;
- beta stage; first integration track — LanAgent for the EBS Lan under a licensing model.
Stack: Python 3.12 · FastAPI · Next.js 15 · React 19 · PostgreSQL + pgvector · Valkey (Redis) · RabbitMQ · OpenSearch · S3/MinIO · gVisor · Helm · Kubernetes · Harbor · GitLab CI