https://github.com/stlimtat/bunshin-go
Bunshin is a go port of LangChain / LangGraph / LangSmith — production-grade LLM pipeline primitives with native concurrency, type safety, and single-binary deploys.
https://github.com/stlimtat/bunshin-go
concurrency go langchain llm llm-agent-framework llm-agentic-workflow
Last synced: 22 days ago
JSON representation
Bunshin is a go port of LangChain / LangGraph / LangSmith — production-grade LLM pipeline primitives with native concurrency, type safety, and single-binary deploys.
- Host: GitHub
- URL: https://github.com/stlimtat/bunshin-go
- Owner: stlimtat
- License: mit
- Created: 2026-06-03T06:46:24.000Z (about 1 month ago)
- Default Branch: master
- Last Pushed: 2026-06-18T11:59:25.000Z (23 days ago)
- Last Synced: 2026-06-18T13:27:56.004Z (23 days ago)
- Topics: concurrency, go, langchain, llm, llm-agent-framework, llm-agentic-workflow
- Language: Go
- Homepage: https://bunshin-go.github.io
- Size: 1.62 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# bunshin-go
[](https://github.com/stlimtat/bunshin-go/actions/workflows/ci.yml)
[](https://pkg.go.dev/github.com/stlimtat/bunshin-go)
[](https://go.dev/doc/install)
[](LICENSE)
A Go port of LangChain / LangGraph / LangSmith — production-grade LLM pipeline primitives with native concurrency, type safety, and single-binary deploys.
| Problem | Python LangChain | bunshin-go |
|---------|-----------------|------------|
| Concurrency | GIL limits parallelism | goroutines, zero overhead |
| Type safety | Runtime errors, brittle | Generics, compile-time checks |
| Deploy | Virtual envs, deps | Single static binary |
| Latency | 50–200 ms startup | Sub-1 ms startup |
| Context windows | 2M tokens = 2 GB RAM | Reference/cursor, O(1) RAM |
**Docs:** [architecture](docs/content/architecture/index.md) · [quickstart](docs/content/quickstart/index.md) · [concepts](docs/content/concepts/index.md) · [how-tos](docs/content/howto/)
---
## Install
```bash
go get github.com/stlimtat/bunshin-go
```
```go
provider := llm.NewFakeProvider("openai", "Hello from bunshin-go!")
result, _ := provider.Complete(ctx, &llm.Request{
Messages: []llm.Message{llm.NewTextMessage(llm.RoleUser, "What is Go?")},
})
fmt.Println(result.Content)
```
---
## CLI
```bash
go build -o bunshin ./cmd/bunshin
```
| Subcommand | What it does |
|------------|--------------|
| `bunshin llm` | Single provider LLM call |
| `bunshin chain` | Two-step entity extraction chain |
| `bunshin agent` | Agent loop with arithmetic tools |
| `bunshin mcp-sandbox` | MCP tool discovery + sandboxed code execution |
| `bunshin serve` | Start the HTTP workflow server |
| `bunshin health` | Self-check (used by Docker healthcheck) |
| `bunshin version` | Print version |
| `bunshin docs` | Generate LLM-ready CLI markdown docs |
```bash
bunshin llm --message "What is Go?"
bunshin agent --question "What is 6*7?"
bunshin serve --addr :9090
```
### Environment variables
| Variable | Default | Description |
|----------|---------|-------------|
| `BUNSHIN_PROVIDER` | `fake` | `fake\|openai\|anthropic\|google\|ollama` |
| `BUNSHIN_MODEL` | _(provider default)_ | Model ID |
| `BUNSHIN_API_KEY` | | API key |
| `BUNSHIN_LOG_LEVEL` | `info` | `debug\|info\|warn\|error` |
| `BUNSHIN_ADDR` | `:8080` | HTTP listen address |
```bash
BUNSHIN_PROVIDER=openai BUNSHIN_API_KEY=sk-... bunshin llm --message "Hello"
```
---
## Docker
```bash
docker compose -f deployments/docker-compose.yml up
curl -X POST http://localhost:8080/workflows/echo \
-H "Content-Type: application/json" \
-d '{"input": {"message": "hello"}}'
curl -N "http://localhost:8080/workflows/echo/stream?input=%7B%22message%22%3A%22hello%22%7D"
```
---
## Testing
```bash
go test -race -count=1 ./...
go test ./pkg/testing/fault/... -v # chaos / fault injection
go test ./pkg/eval/... -v # eval harness
```
---
## Project layout
| Directory | Purpose |
|-----------|---------|
| `pkg/` | Public library packages |
| `internal/` | Private shared utilities |
| `cmd/bunshin/` | Unified CLI (cobra + viper) |
| `api/` | OpenAPI 3.1 specification |
| `deployments/` | Dockerfile, docker-compose |
| `docs/` | Documentation site |
---
## Roadmap
- [ ] OpenAI + Anthropic provider adapters
- [ ] Redis and S3 MessageStore backends
- [ ] gRPC transport
- [ ] LangSmith telemetry backend
- [ ] E2B and Docker sandbox backends
- [ ] Real MCP client (stdio + HTTP/SSE transport)
---
## License
MIT — see [LICENSE](LICENSE).