The PowerToolsyour AI agent reaches for.

Your codebase as a queryable knowledge graph. Built once, read by every tool.

Find anything in your codebase, fast. Five modes, one interface.

Symbol slice, file overview, full file, without burning tokens or re-parsing.

Know what would break before you ship. Three queries against the call graph, one tool.

Your code and your database, on the same interface. 20 actions across six engines.

Twira’s static-analysis (SAST) engine, 54 deterministic detectors across security, bugs, code health, DOM, and port-migration. Drift-resilient suppressions that survive refactors. Local, fast, runs in the agent loop or pre-commit.

Twira’s software composition analysis (SCA) tool. Finds the dependency vulnerabilities that actually reach your code. OSV-backed, reachability-filtered, drift-resilient suppressions. Local, fast, runs in your editor or the agent loop.

Twira's wire-layer PII protection. A local HTTPS reverse proxy that sits between every AI agent on your machine and the LLM provider. Inspects every payload, redacts personal data in flight via reversible token substitution, custodies the provider key (agent never sees it), signs a receipt per call, and writes it to its own offline-verifiable signed ledger. **Included in Pro $29.99/mo, no separate licence.** **Works with your Claude.ai Pro · Max 5x · Max 20x subscription OAuth, ChatGPT Plus · Pro via Codex CLI, AND direct API keys (Anthropic, OpenAI, Google, xAI). Gemini subscription OAuth is the one gap — route Gemini through a Google AI Studio or Vertex API key for now.**

Tamper-evident, cryptographically-signed, append-only record of every meaningful action AI agents take in your codebase. Hunk-level attribution. RFC 3161 timestamps. Offline-verifiable by any third party with no Twira infrastructure. This is the *activity recordkeeping* surface, paired with the [AI Compliance Proxy](/tools/gateway), which handles *content* in flight; together they answer both *what data was handled?* and *can you prove it?*

Four frontier AI brains, one hard question, one peer-reviewed answer. The single best defence against one model hallucinating.

Two Twira sessions on the same repo. No collisions. Coordinated by file claim and broadcast.

Cross-agent institutional memory. Save the lesson once; every future agent, Claude, Codex, Gemini, anything that speaks MCP, sees it before they walk into the same trap.

Type /code-review in your agent. Multiple frontier models review the commit in parallel, each through a different lens. Every confirmed finding is either fixed in this session or tracked as a task, no "we will look at it later."

Type /plan-review in your agent. Multiple frontier models review the implementation plan against your actual codebase, finds reusable code, missing considerations, and architectural issues before you write a single line.

Port a 200,000-line legacy codebase to a new language without losing a single symbol. The deterministic spine your AI agents lean on so they finish what they started, 16 actions, 8 matching tiers, 11 verifier checks, no LLM in the verification path.

The shared project masterplan that EVERY AI agent on EVERY session reads, claims from, and updates. Claude Code, Gemini, Codex, three of each in parallel, all working from the same plan, claiming different tasks, none stepping on each other. One plan; many brains.

Twira’s interactive browser dashboard, running on 127.0.0.1, the visual control surface for everything the agent does, every setting you can change, every event the system emits. Type /ar-dash inside your coding agent (or run twira dashboard open from any terminal) and the dashboard launches in your default browser. 37 pages across four sections, live event stream over WebSocket, signed-cookie auth, 28 backend handlers. The place where developers operate Twira when they do not want to type CLI commands.

Desktop toast banners, audible chime, optional spoken voice readout, and the voice tells you which agent has finished in which session. "Claude Auth bug has finished" in Ryan’s British accent. "Gemini Refactor needs your input" in Sara’s American one. You always know who is asking.