AI-powered CLI for PR automation. Semantic merge conflict resolution, issue-to-code generation, complexity scoring, and full autopilot — all with approval gating.
v0.10.0 · BSL 1.1 · Interactive Overview · Docs
| Command | What happens |
|---|---|
gwi triage <pr> |
Score PR complexity (1–10) |
gwi resolve <pr> |
Resolve merge conflicts semantically |
gwi review <pr> |
Generate review summary |
gwi review --local |
Review staged changes before PR |
gwi issue-to-code <issue> |
Create a PR from a GitHub issue |
gwi autopilot <pr> |
Full pipeline: triage → resolve → review → commit |
gwi gate |
Pre-commit approval gate |
gwi explain <run-id> |
Explain why AI made a decision |
npm install && npm run build# At least one AI provider
export ANTHROPIC_API_KEY="..." # or GOOGLE_AI_API_KEY / OPENAI_API_KEY
export GITHUB_TOKEN="..."gwi triage https://github.com/owner/repo/pull/123
gwi resolve https://github.com/owner/repo/pull/123
gwi issue-to-code https://github.com/owner/repo/issues/456
gwi autopilot https://github.com/owner/repo/pull/123
gwi run approve <run-id>| Tool | What it does | GWI difference |
|---|---|---|
| GitHub Copilot | Suggests code in editor | Generates PRs from issues, resolves conflicts, runs pipelines |
| Cursor / Windsurf | AI coding assistants | Repo-level automation, not editor plugins |
| Linear / Jira | Project management | Actually writes the code, not just tracks it |
| SonarQube | Static analysis | Fixes issues, not just reports them |
Key differentiators: Semantic conflict resolution · Hash-bound approval gating · Multi-agent model routing · Full audit trail · CLI-first
flowchart TB
CLI["gwi CLI"] --> Engine["Workflow Engine"]
Engine --> Agents
Engine --> Storage
Engine --> GH["GitHub API"]
subgraph Agents["AI Agents (8)"]
direction LR
A1["Orchestrator"] & A2["Foreman"] & A3["Triage"] & A4["Coder"]
A5["Resolver"] & A6["Reviewer"] & A7["Slop Detector"] & A8["Infra"]
end
subgraph Storage
Firestore["Firestore (prod)"]
SQLite["SQLite (dev)"]
end
Tasks route to models by complexity:
| Complexity | Model | Use case |
|---|---|---|
| 1–3 | Gemini Flash | Fast scoring, simple triage |
| 4–6 | Claude Sonnet | Code generation, reviews |
| 7–10 | Claude Opus | Complex conflicts, multi-file changes |
stateDiagram-v2
[*] --> Pending: Run created
Pending --> Running
Running --> AwaitingApproval: Patch ready
AwaitingApproval --> Approved: User approves (SHA-bound)
AwaitingApproval --> Rejected
Approved --> Committed: Push to remote
Committed --> [*]
Rejected --> [*]
Running --> Failed
Failed --> [*]
Destructive operations (commit, push, merge) require explicit approval with SHA256 hash binding. If the patch changes after approval, the approval is invalidated.
| Operation | Approval needed? |
|---|---|
| Read / analyze | No |
| Generate patch | No |
| Commit / push | Yes (hash-bound) |
| Merge PR | Yes (hash-bound) |
All agents have SPIFFE identities. Every decision is logged to an immutable audit trail. See SECURITY.md and Threat Model.
git-with-intent/
├── apps/
│ ├── cli/ # gwi command
│ ├── api/ # REST API (Cloud Run)
│ ├── gateway/ # A2A agent coordination
│ ├── github-webhook/ # Webhook handler
│ ├── mcp-server/ # MCP tool server (Cloud Run)
│ ├── worker/ # Background jobs
│ ├── webhook-receiver/ # Generic webhooks
│ ├── registry/ # Workflow templates
│ └── web/ # Dashboard (React)
├── packages/
│ ├── core/ # Storage, billing, security
│ ├── agents/ # AI agent implementations
│ ├── engine/ # Workflow orchestration
│ ├── integrations/ # GitHub/GitLab connectors
│ ├── connectors/ # Airbyte-style connectors
│ ├── forecasting/ # TimeGPT integration
│ ├── sandbox/ # Secure execution
│ └── sdk/ # TypeScript SDK
└── infra/ # OpenTofu (GCP)
flowchart TD
cli["apps/cli"] --> agents["packages/agents"]
cli --> integrations["packages/integrations"]
cli --> core["packages/core"]
agents --> core
integrations --> core
engine["packages/engine"] --> agents
engine --> core
api["apps/api"] --> core
gateway["apps/gateway"] --> core
worker["apps/worker"] --> core
5 Cloud Run services deployed via GitHub Actions + OpenTofu. No direct gcloud deploy.
| Service | Purpose |
|---|---|
gwi-api |
REST API |
gwi-gateway |
A2A agent coordination |
gwi-webhook |
GitHub webhooks |
gwi-worker |
Background jobs |
gwi-mcp-server |
MCP tool server |
Plus Firestore (database) and Firebase Hosting (dashboard).
npm run build # Build all packages
npm run typecheck # Type check
npm run test # Run tests
npm run arv # Agent Readiness Verification (required before commit)20 verification gates run before every commit — forbidden patterns, Zod schema contracts, deterministic golden outputs, security, identity, reliability, observability, and more:
npm run arv:lint # No forbidden patterns
npm run arv:contracts # Zod schema validation
npm run arv:goldens # Deterministic outputs
npm run arv:smoke # Boot checkEvery run produces artifacts at .gwi/runs/<runId>/:
run.json · triage.json · plan.json · patch.diff · review.json · approval.json · audit.log
See CONTRIBUTING.md · Code of Conduct · Security Policy · Support
Business Source License 1.1 — Copyright (c) 2025-2026 Intent Solutions LLC
Source-available. Personal and non-commercial use permitted. Production use in a competing SaaS requires a commercial license. Converts to Apache 2.0 on 2030-02-17. See LICENSE.