Back to directory

Cersei

A Rust SDK that provides every building block of a production coding agent, tool execution, LLM streaming, sub-agent orchestration, persistent memory, skills and MCP integration, as composable library functions rather than a fixed CLI app.

Cersei

Cersei exposes an Agent::builder() API that composes a model provider, a set of tools, a permission policy and a prompt into a single asynchronous call, so an agent can be embedded directly in another application rather than run only as a standalone CLI. It supports multiple providers, unlike a single-provider agent, and stores memory as files plus a graph rather than only flat files or only SQLite.

Abstract, a companion CLI built on Cersei, ships as one binary with zero runtime dependencies and graph memory by default, and its benchmark against Claude Code reports an 8.2 times faster warm startup (32ms versus 266ms), a binary about 29 times smaller (6.0MB versus 174MB) and around 68 times less resident memory (4.9MB versus 333MB). It installs with cargo install --path crates/abstract-cli and supports an interactive REPL, a single-shot prompt, session resume, an Opus-with-max-thinking mode, and a JSON, no-permissions CI mode.

It reads its own tool and skill formats as well as those used by Claude Code and OpenCode. It is released under the MIT License.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.