cersei (Pacifio)

瑟曦

用于构建编码代理的完整Rust SDK。

Cersei为您提供了生产编码代理的每个构建块——工具执行、LLM流、子代理编排、持久内存、技能、MCP集成——作为可组合库函数。构建一个克劳德代码替代品，在你的应用程序中嵌入一个代理，或者创建一些全新的东西。

use cersei::prelude::*;

#[tokio::main]
async fn main() -> anyhow::Result {
    let output = Agent::builder()
        .provider(Anthropic::from_env()?)
        .tools(cersei::tools::coding())
        .permission_policy(AllowAll)
        .run_with("Fix the failing tests in src/")
        .await?;

    println!("{}", output.text());
    Ok(())
}

______________________________________________________________________

为什么选择瑟曦

Cersei基于Claude Code（逆向工程Rust端口）的架构构建，旨在让任何人都可以构建一个完整的、可直接替换的Claude Code、OpenCode或任何编码代理——作为库调用。

______________________________________________________________________

摘要——CLI

摘要 是基于Cersei构建的完整CLI编码代理。一个二进制文件，零运行时依赖关系，默认情况下使用图形内存。

# Install
cargo install --path crates/abstract-cli

# Use
abstract                           # Interactive REPL
abstract "fix the failing tests"   # Single-shot
abstract --resume                  # Resume last session
abstract --model opus --max        # Opus with max thinking
abstract --no-permissions --json   # CI mode with NDJSON output

抽象代码与克劳德代码

所有号码来自 run_tool_bench.sh --full.

Claude Code的记忆回忆每次都会调用Sonnet，根据相关性（测量7.5秒）对前5个文件进行排名。 Abstract的图在98微秒内完成索引查找——同样的功能，无需LLM调用，无需API成本。

完整基准： crates/abstract-cli/benchmarks/REPORT.md

特性

• 34个内置工具（文件、shell、web、规划、编排、调度）
• 多提供商：Anthropic+OpenAI（+Ollama、Azure、vLLM）
• 默认情况下，图形内存（Grafeo）处于打开状态
• 汽车紧凑型、汽车梦想、努力水平（低/中/高/最大）
• MCP服务器支持
• 会话持久性（兼容Claude Code的JSONL）
• 具有会话缓存的交互式权限
• 12个斜线命令(/help, /commit, /review, /memory, /model, /diff等等）
• 带有语法高亮显示的流式markdown渲染
• TOML配置： ~/.abstract/config.toml + .abstract/config.toml
• 管道的JSON输出模式(--json)

______________________________________________________________________

安装

[dependencies]
cersei = { git = "https://github.com/pacifio/cersei" }
tokio = { version = "1", features = ["full"] }
anyhow = "1"

对于图形备份内存（可选）：

cersei-memory = { git = "https://github.com/pacifio/cersei", features = ["graph"] }

______________________________________________________________________

建筑

cersei                    Facade crate — use cersei::prelude::*;
  cersei-types            Provider-agnostic messages, errors, stream events
  cersei-provider         Provider trait + Anthropic/OpenAI implementations
  cersei-tools            30+ tools, permissions, bash classifier, skills, git utils
  cersei-tools-derive     #[derive(Tool)] proc macro
  cersei-agent            Agent builder, agentic loop, compact, coordinator, effort
  cersei-memory           Memory trait, memdir, CLAUDE.md, sessions, Grafeo graph
  cersei-hooks            Hook/middleware system
  cersei-mcp              MCP client (JSON-RPC 2.0, stdio transport)
abstract-cli              CLI coding agent ("abstract") — REPL, commands, config, permissions

______________________________________________________________________

核心概念

提供者

任何LLM后端。内置：Anthropic（使用OAuth）、OpenAI（与Ollama、Azure、vLLM兼容）。

Agent::builder().provider(Anthropic::from_env()?)           // Anthropic API key
Agent::builder().provider(OpenAi::builder()
    .base_url("http://localhost:11434/v1")                   // Ollama
    .model("llama3.1:70b").api_key("ollama").build()?)
Agent::builder().provider(MyCustomProvider)                  // impl Provider

工具（30+）

编码代理所需的每个工具，按组组织：

cersei::tools::all()           // 30+ tools
cersei::tools::coding()        // filesystem + shell + web
cersei::tools::filesystem()    // Read, Write, Edit, Glob, Grep, NotebookEdit
cersei::tools::shell()         // Bash, PowerShell
cersei::tools::web()           // WebFetch, WebSearch
cersei::tools::planning()      // EnterPlanMode, ExitPlanMode, TodoWrite
cersei::tools::scheduling()    // CronCreate/List/Delete, Sleep, RemoteTrigger
cersei::tools::orchestration() // SendMessage, Tasks (6 tools), Worktree

10行自定义工具：

#[derive(Tool)]
#[tool(name = "search", description = "Search docs", permission = "read_only")]
struct SearchTool;

#[async_trait]
impl ToolExecute for SearchTool {
    type Input = SearchInput; // derives Deserialize + JsonSchema
    async fn run(&self, input: SearchInput, ctx: &ToolContext) -> ToolResult {
        ToolResult::success(format!("Found: {}", input.query))
    }
}

子代理编排

生成并行工作者，协调任务，在代理之间传递消息：

// AgentTool — model spawns sub-agents autonomously
Agent::builder()
    .tool(AgentTool::new(|| Box::new(Anthropic::from_env()?), cersei::tools::coding()))

// Coordinator mode — orchestrate parallel workers
Agent::builder()
    .tools(cersei::tools::all())  // includes Agent, Tasks, SendMessage
    // Workers get filtered tools (no Agent — prevents recursion)

// Task system
// TaskCreate → TaskUpdate → TaskGet → TaskList → TaskStop → TaskOutput

内存（三层）

use cersei::memory::manager::MemoryManager;

let mm = MemoryManager::new(project_root)
    .with_graph(Path::new("./memory.grafeo"))?;  // optional graph layer

// Tier 1: Flat files (~/.claude/projects//memory/)
let metas = mm.scan();                    // scan .md files with frontmatter
let content = mm.build_context();         // build system prompt injection

// Tier 2: CLAUDE.md hierarchy (managed > user > project > local)
// Automatically merged into build_context()

// Tier 3: Graph memory (Grafeo, optional)
let id = mm.store_memory("User prefers Rust", MemoryType::User, 0.9)?;
mm.tag_memory(&id, "preferences");
let results = mm.recall("Rust", 5);       // graph query with fallback to text match

// Session persistence (JSONL, append-only, tombstone soft-delete)
mm.write_user_message("session-1", Message::user("Hello"))?;
let messages = mm.load_session_messages("session-1")?;

技能（克劳德代码+兼容OpenCode）

// Auto-discovers skills from:
//   .claude/commands/*.md      (Claude Code format)
//   .claude/skills/*/SKILL.md  (OpenCode format)
//   ~/.claude/commands/*.md    (user-level)
//   Bundled skills             (simplify, debug, commit, verify, stuck, remember, loop)

let skill_tool = SkillTool::new().with_project_root(".");
// skill="list" → lists all available skills
// skill="debug" args="tests are flaky" → expands $ARGUMENTS template

实时事件

三种观察机制：

// 1. Callback
Agent::builder().on_event(|e| match e {
    AgentEvent::TextDelta(t) => print!("{}", t),
    AgentEvent::ToolStart { name, .. } => eprintln!("[{}]", name),
    _ => {}
})

// 2. Broadcast (multi-consumer)
let agent = Agent::builder().enable_broadcast(256).build()?;
let mut rx = agent.subscribe().unwrap();
tokio::spawn(async move { while let Ok(e) = rx.recv().await { /* ... */ } });

// 3. Stream (bidirectional control)
let mut stream = agent.run_stream("Deploy");
while let Some(e) = stream.next().await {
    if let AgentEvent::PermissionRequired(req) = e {
        stream.respond_permission(req.id, PermissionDecision::Allow);
    }
}

上下文管理

Agent::builder()
    .auto_compact(true)          // summarize old messages at 90% context usage
    .compact_threshold(0.9)      // trigger threshold
    .tool_result_budget(50_000)  // truncate oldest tool results above 50K chars
    .thinking_budget(8192)       // extended thinking tokens
    .effort(EffortLevel::High)   // Low/Medium/High/Max

MCP（模型上下文协议）

let mcp = McpManager::connect(&[
    McpServerConfig::stdio("db", "npx", &["-y", "@my/db-mcp"]),
    McpServerConfig::sse("docs", "https://mcp.example.com"),
]).await?;

Agent::builder().tools(mcp.tool_definitions().await)

OAuth（人类原生）

// Opens browser, PKCE flow, token storage, refresh
cargo run --example oauth_login

______________________________________________________________________

代理生成器-完整的API

Agent::builder()
    // Provider (required)
    .provider(Anthropic::from_env()?)

    // Tools
    .tool(MyTool)
    .tools(cersei::tools::coding())

    // Model & generation
    .model("claude-sonnet-4-6")
    .max_turns(10)
    .max_tokens(16384)
    .temperature(0.7)
    .thinking_budget(8192)

    // Prompt
    .system_prompt("You are a helpful assistant.")
    .append_system_prompt("Extra context.")

    // Environment
    .working_dir("./my-project")
    .permission_policy(AllowAll)          // or AllowReadOnly, DenyAll, RuleBased, Interactive

    // Memory
    .memory(JsonlMemory::new("./sessions"))
    .session_id("my-session")

    // Hooks & events
    .hook(CostGuard { max_usd: 5.0 })
    .on_event(|e| { /* ... */ })
    .enable_broadcast(256)
    .reporter(ConsoleReporter { verbose: true })

    // Context management
    .auto_compact(true)
    .compact_threshold(0.9)
    .tool_result_budget(50_000)

    // Execute
    .build()?                             // -> Agent
    .run_with("Fix the tests")            // -> AgentOutput (shorthand)

______________________________________________________________________

基准测试

在Apple Silicon上测量，发布版本，100次迭代，3次预热运行。

工具输入/输出

vs克劳德代码CLI

注： Cersei是一个库——工具调度发生在进程中。Claude Code是一个CLI，其中 每个子代理分支支付全部启动成本。这些是不同的层次；下面的比较 显示了进程内调度和CLI进程开销之间的差距。

有关苹果对苹果CLI的比较，请参阅 抽象CLI基准测试.

内存I/O

基准套房

每个工作台都位于自己的独立目录中，有自己的运行器和结果模式。添加新的长椅作为兄弟姐妹。

运行基准测试

# Rust-side SDK benches (no external services)
cargo run --example benchmark_io --release
cargo run --release -p abstract-cli --example memory_bench

# vs Claude Code / Codex CLIs
./run_tool_bench.sh --iterations 20 --full

# Python-harness benches (uv-managed; each dir self-contained)
./bench/general-agents/run.sh          # Cersei vs Agno / PydanticAI / LangGraph / CrewAI
./bench/term-bench/run.sh              # Terminal Bench 2.0 via Daytona

# LongMemEval memory benchmark (head-to-head vs Mastra / Zep / Supermemory)
./bench/long-mem/setup.sh              # downloads oracle + s datasets
OPENAI_API_KEY=sk-… cargo run --release -p longmem-bench -- \
  --dataset s --config all --concurrency 8

# Real-LLM compression savings (requires API keys)
OPENAI_API_KEY=sk-… cargo test -p cersei-agent \
  --test e2e_openai_compression -- --ignored --nocapture

______________________________________________________________________

压力测试

cargo run --example stress_core_infrastructure --release  # system prompt, compact, context, bash classifier
cargo run --example stress_tools --release                 # all 30+ tools, registry, performance
cargo run --example stress_orchestration --release         # sub-agents, coordinator, tasks, messaging
cargo run --example stress_skills --release                # bundled + disk skills, Claude Code + OpenCode format
cargo run --example stress_memory --release                # memdir, CLAUDE.md, sessions, extraction, auto-dream

______________________________________________________________________

示例

cargo run --example simple_agent --release

______________________________________________________________________

测试套件

# Run all 160 unit tests
cargo test --workspace

# Run with graph memory (requires grafeo)
cargo test --workspace --features graph

# Run specific crate
cargo test -p cersei-tools
cargo test -p cersei-agent
cargo test -p cersei-memory
cargo test -p cersei-mcp

160个单元测试 | 262次压力检查 | 0次失败 | 零I/O回归

______________________________________________________________________

扩展点

______________________________________________________________________

文档

cersei.pacifio.dev/docs -提供API参考、架构、食谱、基准测试和llms.txt支持的完整文档。

______________________________________________________________________

许可证

MIT许可证

版权所有（c）2025 Adib Mohsin

特此免费授予任何获得本软件和相关文档文件（“软件”）副本的人在不受限制的情况下处理软件的权限，包括但不限于使用、复制、修改、合并、发布、分发、再许可和/或销售软件副本的权利，以及允许获得软件的人这样做，但须符合以下条件：

上述版权声明和本许可声明应包含在软件的所有副本或实质部分中。

软件按“原样”提供，不提供任何明示或暗示的保证，包括但不限于适销性、特定用途适用性和非侵权性的保证。在任何情况下，作者或版权持有人均不对因软件或软件的使用或其他交易而产生或与之相关的任何索赔、损害赔偿或其他责任承担责任，无论是在合同、侵权或其他诉讼中。