nexssp/flow
Go runtime and language for composing .nflow programs from registered actions. The nflow CLI runs flows, lints them, builds standalone binaries, and inspects the active compiler surface. Built on nexssp/kernel.
- Pipeline operators: ->, ||, &, projection { ... }
- Modifiers attach Kernel resilience: :timeout= :retry= :cache= :auth
- Directives: @require @include @assert @on_error @config:strict
- Standalone binaries via `nflow build -o FILE flow.nflow`
- Microkernel: core.Bundle is the only contract extensions implement
- Language-agnostic @require - Python, Rust, Node, C++ all welcome
nexssp/srcpack
Source code compressor & context engineering engine built for LLMs, Claude Desktop, Cursor, Windsurf, MCP, and A2A multi-agent DAG pipelines.
- AST token compression (65%–85% context reduction)
- Multi-transport: CLI, MCP Server (stdio), REST API & A2A
- In-memory secret masking & visual ASCII token tree index
- Multi-language: Go, TypeScript, TSX, Vue, Python, Rust, Zig
nexssp/kernel
Lightweight, typed Go execution library for reusable actions, resilient pipelines, and deterministic DAG workflows. Independent of transports, persistence, AI vendor SDKs, and identity providers.
- action: typed builders, hooks, middleware, composition, retries, sagas
- observe: vendor-neutral events, slog, Prometheus, JSONL, Memory sinks
- ai/dag: typed DAGs, Kahn-cycle-detect, pooled state, Mermaid export, HIL
- xctx: pooled RequestScope; xerr: 11 categorized error kinds
- stream: zero-alloc iter.Seq2 operators (Map, Filter, Take, Batch, Reduce)
- ringbuf: cache-line-aligned lock-free MPMC ring buffer
nexssp/transportai
Turn existing business actions directly into AI Agent tools for Anthropic Claude, Cursor, Windsurf, and multi-agent coordination frameworks.
- tmcp: Model Context Protocol (MCP) over stdio & SSE
- ta2a: Agent2Agent protocol with Agent Cards & task states
- Automated tool schema generation from Go structs
- Dynamic prompt templates with variable validation
nexssp/transport
Bind kernel actions simultaneously to HTTP/REST, Server-Sent Events, CLI subcommands, in-process event bus, cron schedules, and priority worker queues.
- thttp: REST, SSE, streaming NDJSON & tag binding
- tcli: Flag parsing (cli:"flag,f"), args & auto-generated help
- cron & tworker: Scheduled jobs & background workers
- taskqueue: High/Normal/Low priority lanes with DLQ
nexssp/testkit
Zero-dependency integration and performance testing suite built for high-throughput systems, bypassing heavy Docker environments.
- RunSmokeTests: Automated endpoint fuzzing to catch panics
- Simulate: Concurrency & thundering herd synchronization
- LoadTest: In-memory RPS, p50/p95/p99 latency benchmarking
- ListenSSE: Real-time Server-Sent Event stream testing
Architecture Evidence
Real code from the open repositories
These snippets are pulled directly from the nexssp/kernel README and nexssp/flow example .nflow files. Clone either repo and run them as-is.
// 1. One-line CLI context generation
// $ srcpack arch --tree -c
// Extracts exported AST signatures, prepends ASCII token tree, masks secrets, copies to clipboard.
// 2. Mount as an MCP Tool for Claude / Cursor / Windsurf
server := tmcp.New("srcpack-mcp", "1.1.0")
server.Mount(actions.AllActions)
_ = server.ServeStdio(ctx)
// 3. Multi-Agent DAG Pipeline (Agent-to-Agent)
client := ta2a.NewClient("http://localhost:8091")
packerAct := client.AsAction("remote.packer", "packer").Build()
pipeline, _ := dag.New("ai_review_dag").
AddNode("pack_ctx", "context", packerAct).
AddNode("llm_audit", "audit_report", llmAuditAct).
AddEdge("pack_ctx", "llm_audit").
Compile()All execution pipelines compile into zero-allocation call graphs at Build() time. Context authorization, idempotency, distributed locks, and sagas execute with strict Go idioms.
Next step
The foundation is open. Senior systems delivery is our craft.
Use the open-source packages directly, or partner with NEXSS to build, scale, and modernize your mission-critical software systems.