v1.0 · Live Project-Personalization · Feature-Driven · Multi-Agent

The Agent That Cooks Your |

Start with a generic project. End with a perfectly tailored solution. OpenCook autonomously implements precise, production-grade features to personalize any standard codebase. Driven by deep function characterization and orchestrated subagents, it seamlessly injects custom functionalities exactly where you need them.

Start Cooking Explore Personalization

6Recipe Tiers

4Memory Layers

∞LLM Providers

Claude Code

Codex

OpenClaw

OpenCode

TRAE

Gemini CLI

Connecting to Vimeo

Loading the demo video...

Video temporarily unavailable

Please check your network status. If Vimeo is blocked, use the source link below.

Source video link: https://vimeo.com/1178451557 Open Source Video

Live Session

opencook — bash

❯ opencook --target postgresql --func array_to_string

[PlanAgent] Scanning /src/backend/utils/adt/varlena.c ...

[Memory] Loading episodic context · 12 prior tasks found

[CodeAgent] Located PG_FUNCTION_INFO_V1 registration site

[CodeAgent] Writing · builtins.c · pg_proc.dat · sql/functions.sql

✔ array_to_string — implementation complete

[TestAgent] Running pg_regress · 48 / 48 passed

✔ Patch ready · 3 files · 0 regressions

Competitive Landscape

How OpenCook
Stands Apart

Project-local recipes, staged role separation, per-agent model routing, and live-verified patches for project-specific personalization built from the ground up.

Personalization Demo — Coming Soon

Walkthrough: OpenCook personalizing a database codebase by adding a custom production-ready function from task input to verified patch

Code-Observable Surface	OpenCook	Claude Code	Codex	OpenCode	OpenClaw
Personalization Mechanism	✦ Walk-up recipe roots + wisdom self-accumulates across tasks	~ Global skills; no cross-task learning	~ Static file config	~ Predefined agents	~ Pre-built integrations
Delivery Loop Architecture	✦ Role-separated staged pipeline	~ No stage separation	~ Dynamic routing	~ No inter-stage handoff	~ Gateway dispatch
Patch-Oriented Traceability	✦ Trajectory + patch + HTML report	~ No patch artifact	~ Thread log only	~ No change record	~ No patch traceability
In-Tree Scaffold + Live Verification	✦ In-tree conventions + live execution verification	✗	✗	✗	✗
Per-Agent Role Isolation	✦ Independent model + tool scope per role	~ No per-role model binding	✗	✗	✗

✦ Clearly present in inspected source ~ Present but narrower or less explicit ✗ Not observed in inspected source

Personalization Success Cases

Function Intent In,
Personalization Out

One verified OpenCook case: tailoring a database codebase by implementing custom functions and producing patch-ready results.

Source Context

How It Works

Built Different,
Built for Personalization

Three specialized agents, six capabilities, and deep project context for tailoring real codebases with production-grade features.

Working Episodic Project Long-Term

Task InputCLI · Shell · Script

→

Read-onlyPlanAgentRepository scoping

→

OrchestratorCodeAgentImplements · Coordinates

→

ValidatorTestAgentVerifies patch

→

Verified PatchTrajectory · Diff

Four-Layer Memory

Working → Episodic → Project → Long-Term across sessions.

Provider-Agnostic Models

Anthropic, OpenAI, Doubao, ModelScope — swap per agent independently.

Subagent Orchestration

PlanAgent scopes read-only. CodeAgent implements. TestAgent verifies.

Dynamic Recipes

Drop a recipe folder with a SKILL.md entry file on the path — loaded at runtime.

Context-Aware Personalization

Deep understanding of cross-unit dependencies and complex build conventions to seamlessly inject custom features into any generic project.

Progressive Validation

Syntax checking, compliance verification, and semantic testing to ensure your personalized features achieve production-grade reliability.

Get Started

Up & Running
in Minutes

Four steps to give any coding agent project-specific recipes, rules, and memory — and start personalizing your codebase.

Clone

bash

git clone https://github.com/weAIDB/OpenCook.git
cd OpenCook

Install

bash

uv venv --python 3.10
source .venv/bin/activate
uv pip install -e .

Configure

yaml

# Target database
databases:
  default: sqlite  # postgresql | duckdb

# LLM provider & API key
model_providers:
  anthropic:
    api_key: your_api_key_here

Cook

bash

# Interactive TUI
opencook interactive \
  --config-file opencook_config.yaml

# Headless
opencook run "Add BOOL_AND for SQLite" \
  --config-file opencook_config.yaml

The Agent That Cooks Your |

How OpenCookStands Apart

Function Intent In,Personalization Out

Built Different,Built for Personalization

Four-Layer Memory

Provider-Agnostic Models

Subagent Orchestration

Dynamic Recipes

Context-Aware Personalization

Progressive Validation

Up & Runningin Minutes

Ready to Cook?

How OpenCook
Stands Apart

Function Intent In,
Personalization Out

Built Different,
Built for Personalization

Up & Running
in Minutes