systemprompt-template

<div align="center"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://systemprompt.io/files/images/logo.svg"> <source media="(prefers-color-scheme: light)" srcset="https://systemprompt.io/files/images/logo-dark.svg"> <img src="https://systemprompt.io/files/images/logo-dark.svg" alt="systemprompt.io" width="380"> </picture> # Own how your organization uses AI. ### The narrow waist between your AI and everything it touches. Self-hosted. Air-gapped. Every interaction governed and provable. [](https://github.com/systempromptio/systemprompt-core) [](LICENSE) [](https://github.com/systempromptio/systemprompt-core/blob/main/LICENSE) [](https://www.rust-lang.org/) [](https://www.postgresql.org/) [**systemprompt.io**](https://systemprompt.io) · [**Documentation**](https://systemprompt.io/documentation/) · [**Guides**](https://systemprompt.io/guides) · [**Discord**](https://discord.gg/wkAbSuPWpr) </div> --- ## This is the local evaluation **systemprompt.io is AI Governance Infrastructure** — the governance layer for AI agents. A single compiled Rust binary that authenticates, authorises, rate-limits, logs, and costs every AI interaction. Self-hosted, air-gap capable, provider-agnostic. This repo is how you evaluate it. Clone it, run it on your own machine, bring your own AI key, and watch every claim below execute against 40+ scripted demos. **The terminal recordings are real captures of those demos running — not mockups.** Regenerate them yourself with `just record-svgs`. > Template: MIT. [systemprompt-core](https://github.com/systempromptio/systemprompt-core): BSL-1.1 — free for evaluation and non-production use. Production requires a commercial license. --- ## Evaluate it in three commands ```bash just build # 1. compile the workspace just setup-local <anthropic> <openai> <gemini> # 2. profile + Postgres + publish just start # 3. serve governance, agents, MCP, admin, API ``` Open **http://localhost:8080** and run `systemprompt --help`. Point Claude Code, Claude Desktop, or any MCP client at it — permissions follow the user, not the client. **Prerequisites.** Rust 1.75+ · [`just`](https://just.systems) · Docker · `jq` · `yq` · one of an Anthropic, OpenAI, or Gemini API key · free ports `8080` and `5432`. Running a second clone side-by-side? `just setup-local <key> "" "" 8081 5433`. --- ## Govern. Every tool call. Four layers of synchronous in-process enforcement on every tool call: **scope** check → **secret** scan → **block**list → **rate** limit. No sidecar. No proxy. Deny reasons are structured and auditable. ### Four-layer governance pipeline <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/01-governance.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/01-governance.svg"> <img src="demo/recording/svg/output/dark/01-governance.svg" alt="Governance pipeline — terminal recording" width="820"> </picture> > Every tool call hits `PreToolUse → govern`. Scope, secret scan, blocklist, rate limit — all synchronous, all in-process, all audited. → [`demo/recording/svg/svg-01-governance.sh`](demo/recording/svg/svg-01-governance.sh) ### Secrets never touch inference <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/02-secrets.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/02-secrets.svg"> <img src="demo/recording/svg/output/dark/02-secrets.svg" alt="Secrets management — terminal recording" width="820"> </picture> > Even admin-scope agents can't leak an AWS key, a GitHub PAT, or an RSA private key. Credentials are injected server-side at tool-call time. → [`svg-02-secrets.sh`](demo/recording/svg/svg-02-secrets.sh) ### Thousands of governed requests per second <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/06-benchmark.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/06-benchmark.svg"> <img src="demo/recording/svg/output/dark/06-benchmark.svg" alt="Governance benchmark — terminal recording" width="820"> </picture> > Measure it yourself with `just benchmark`. Author's laptop: **3,308 req/s** on the burst run, p50 13.5 ms / p99 22.7 ms — every request running JWT validation, scope resolution, three rule evaluations, and an async audit write. Governance adds <1% to AI response time. → [`svg-06-benchmark.sh`](demo/recording/svg/svg-06-benchmark.sh) --- ## Prove. Every decision. Structured evidence for every interaction. Every tool call produces a five-point audit trace — the shape auditors ask for, not the shape your logging framework happens to emit. ### Queryable audit trail <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/03-audit.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/03-audit.svg"> <img src="demo/recording/svg/output/dark/03-audit.svg" alt="Audit trail — terminal recording" width="820"> </picture> > Identity → Agent Context → Permissions → Tool Execution → Result. Queryable, SIEM-ready, cost-attributed. → [`svg-03-audit.sh`](demo/recording/svg/svg-03-audit.sh) ### Replay any request, end to end <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/05-tracing.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/05-tracing.svg"> <img src="demo/recording/svg/output/dark/05-tracing.svg" alt="Request tracing — terminal recording" width="820"> </picture> > Reconstruct any request as the model, the permission engine, and the operator saw it. → [`svg-05-tracing.sh`](demo/recording/svg/svg-05-tracing.sh) --- ## Standardise. Every team. One CLI, one binary, one database — the same surface local and remote. Register agents, distribute skills by role, run them under identical policy everywhere. ### Multi-agent orchestration <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/07-agent.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/07-agent.svg"> <img src="demo/recording/svg/output/dark/07-agent.svg" alt="Agent orchestration — terminal recording" width="820"> </picture> > Discover, configure, message, trace, and register agents. MCP and A2A at the protocol layer. → [`svg-07-agent.sh`](demo/recording/svg/svg-07-agent.sh) ### MCP servers under central governance <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/04-mcp-tracking.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/04-mcp-tracking.svg"> <img src="demo/recording/svg/output/dark/04-mcp-tracking.svg" alt="MCP tool tracking — terminal recording" width="820"> </picture> > Central registry, per-server OAuth2, scoped tool exposure, end-to-end access logs. → [`svg-04-mcp-tracking.sh`](demo/recording/svg/svg-04-mcp-tracking.sh) ### One binary, every capability compiled in <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/08-skills.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/08-skills.svg"> <img src="demo/recording/svg/output/dark/08-skills.svg" alt="Extensions and capabilities — terminal recording" width="820"> </picture> > 12 extensions, 71 sqlx-checked schemas, 13 jobs — all in one process, all discoverable via the CLI. → [`svg-08-skills.sh`](demo/recording/svg/svg-08-skills.sh) ### Services, database, and logs in one CLI <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/09-infrastructure.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/09-infrastructure.svg"> <img src="demo/recording/svg/output/dark/09-infrastructure.svg" alt="Infrastructure overview — terminal recording" width="820"> </picture> > `infra services status` → `infra db status` → `infra logs view`. No sidecars, no sprawl. → [`svg-09-infrastructure.sh`](demo/recording/svg/svg-09-infrastructure.sh) ### Identity, roles, and network-layer bans <picture> <source media="(prefers-color-scheme: dark)" srcset="demo/recording/svg/output/dark/10-users.svg"> <source media="(prefers-color-scheme: light)" srcset="demo/recording/svg/output/light/10-users.svg"> <img src="demo/recording/svg/output/dark/10-users.svg" alt="Users and identity — terminal recording" width="820"> </picture> > Users are first-class; roles scope every tool call; IP bans are audited. Governance follows the user, not the client. → [`svg-10-users.sh`](demo/recording/svg/svg-10-users.sh) --- ## How it's built One language. One database. One binary. One CLI. - **Rust workspace** — `core/` is a read-only BSL-1.1 submodule; your code lives in `extensions/`. - **PostgreSQL 18+** — the only runtime dependency. Air-gap deploys are a `docker run` away. - **Thousands of req/s** — the pipeline is synchronous and in-process, not a sidecar. 3.3k req/s on a laptop, p50 ~13 ms. - **Provider-agnostic** — Anthropic, OpenAI, Gemini — swap at the profile level. - **MCP · A2A · OAuth2 · WebAuthn** — governed at the protocol layer, not bolted on. ``` my-eval/ ├── extensions/ # Your Rust code (compile-time extensions) ├── services/ # Config-only (YAML/Markdown): agents, skills, plugins, providers ├── demo/ # 40+ runnable evaluation scripts + recording pipeline ├── storage/files/ # Static assets (CSS, JS, images) ├── Cargo.toml # Workspace manifest ├── justfile # Development commands └── CLAUDE.md # AI assistant instructions ``` --- ## Demo index 40+ executable evaluation scripts across ten categories. Each script is numbered — run them in order. | Category | Scripts | Exercises | Recording | |---|---|---|---| | [`demo/governance/`](demo/governance/) | 8 | Tool-call approvals, denials, secret breach, rate limits, hooks | ✓ `01`, `02` | | [`demo/agents/`](demo/agents/) | 5 | Agent lifecycle, config, messaging, tracing, A2A registry | ✓ `07` | | [`demo/mcp/`](demo/mcp/) | 3 | MCP servers, access tracking, tool execution | ✓ `04` | | [`demo/skills/`](demo/skills/) | 5 | Skills, content, files, plugins, contexts | ✓ `08` | | [`demo/infrastructure/`](demo/infrastructure/) | 5 | Services, database, jobs, logs, config | ✓ `09` | | [`demo/analytics/`](demo/analytics/) | 8 | Overview, agents, costs, requests, sessions, content/traffic, conversations, tools | ✓ `03` | | [`demo/users/`](demo/users/) | 4 | User CRUD, roles, sessions, IP bans | ✓ `10` | | [`demo/performance/`](demo/performance/) | 2 | Request tracing, benchmarks, load testing | ✓ `05`, `06` | | [`demo/web/`](demo/web/) | 2 | Content types, templates, sitemaps, validation | – | | [`demo/cloud/`](demo/cloud/) | 1 | Auth, profiles, deployment info | – | See [`demo/README.md`](demo/README.md) for the full catalogue, [`demo/AGENTS.md`](demo/AGENTS.md) for the LLM-targeted runbook, and [`demo/recording/svg/README.md`](demo/recording/svg/README.md) for how to regenerate the terminal SVGs above. --- ## Reference | `just` target | Description | |---|---| | `just build` | Build the workspace | | `just setup-local [keys] [http_port] [pg_port]` | Seed local profile, start Docker Postgres, run publish pipeline | | `just start` | Start all services | | `just publish` | Compile templates, bundle CSS/JS, copy assets | | `just record-svgs [N…]` | Regenerate the terminal SVGs above | | `just db-up` / `db-down` / `db-reset` | Manage the local Postgres container | | `just clippy` | Lint the workspace (pedantic, deny-all) | `systemprompt --help` is the entry point for the CLI. Every domain — `core`, `infra`, `admin`, `cloud`, `analytics`, `web`, `plugins`, `build` — is discoverable with `systemprompt <domain> --help`. --- ## License & production use **This template** is MIT. Fork it, modify it, use it however you like — **for local evaluation**. **[systemprompt-core](https://github.com/systempromptio/systemprompt-core)** is [BSL-1.1](https://github.com/systempromptio/systemprompt-core/blob/main/LICENSE): free for evaluation, testing, and non-production use. **Production use requires a commercial license.** Each version converts to Apache 2.0 four years after publication. **Evaluate Free** → you're already here. **Book a Meeting** → [**systemprompt.io**](https://systemprompt.io) · [ed@systemprompt.io](mailto:ed@systemprompt.io) --- <div align="center"> **[systemprompt.io](https://systemprompt.io)** · **[systemprompt-core](https://github.com/systempromptio/systemprompt-core)** · **[Documentation](https://systemprompt.io/documentation/)** · **[Guides](https://systemprompt.io/guides)** · **[Discord](https://discord.gg/wkAbSuPWpr)** _{Own how your organization uses AI. Every interaction governed and provable.} </div>