Introducing Weave
A harness-agnostic multi-agent orchestration framework that brings structure, specialization, and safety to AI-assisted software development.
We built Weave because single-agent workflows were breaking down under the weight of real-world software engineering tasks. When one agent is responsible for planning, implementing, reviewing, and security auditing, quality suffers and risks multiply.
Weave is a harness-agnostic multi-agent orchestration framework that coordinates specialized agents through a structured workflow with mandatory review gates.
Why Weave Exists
Modern AI coding assistants excel at focused tasks but struggle when asked to juggle multiple responsibilities. A single agent cannot effectively:
- Plan a complex feature while simultaneously implementing it
- Write code while maintaining critical distance for security review
- Balance velocity with thoroughness across planning, execution, and verification
The result is inconsistent quality, missed security issues, and plans that drift from implementation.
Weave solves this by separating concerns across eight specialized agents, each with a distinct role and clear boundaries.
What Makes Weave Different
DSL-First Configuration
Weave workflows are defined in .weave files using a declarative DSL. This makes orchestration logic explicit, versionable, and portable across different AI harnesses.
Adapter Pattern for Portability
Weave is not tied to a single AI platform. It uses an adapter pattern to work across multiple harnesses, allowing teams to choose the environment that fits their workflow.
Eight Specialized Agents
Each agent has a focused responsibility:
- Planner: Breaks down requirements into structured task plans
- Implementer: Executes individual tasks from the plan
- Reviewer: Performs code review with quality and security focus
- Security Auditor: Dedicated security analysis and threat modeling
- Tester: Writes and maintains test coverage
- Documenter: Maintains technical documentation
- Researcher: Investigates technical questions and options
- Coordinator: Orchestrates workflow and enforces gates
Mandatory Review Gates
Weave enforces review checkpoints through two mechanisms:
- Weft: Pre-commit review gate that blocks merging until review passes
- Warp: Post-commit audit trail for compliance and retrospective analysis
These gates ensure that code is reviewed before it ships and that the review process is auditable.
Supported Harnesses
Weave currently supports three AI coding environments:
- OpenCode: Open-source AI coding assistant
- Pi: Inflection AI’s assistant platform
- Claude Code: Anthropic’s coding-focused interface
The adapter architecture allows new harnesses to be added without changing workflow definitions.
Getting Started
Ready to try Weave? Head over to the overview documentation to learn how to install Weave, write your first workflow, and configure agents for your project.
Weave is designed for teams that need structure, safety, and specialization in their AI-assisted development workflow.