OpenAI builds and deploys advanced AI models like GPT-4o for autonomous agents and workflows.
Iris
FreeStarting price
0Popularity
About Iris
Iris operates as a local proxy between Claude Code and Anthropic’s API, listening on 127.0.0.1:8787. It intercepts every request Claude Code makes, inspecting the payload to reveal token composition, tool schemas, and conversation history. The tool evaluates which tools are actually used versus those merely present in context, allowing users to disable unused schemas to reduce token load. Iris also enforces execution policies before tool calls run, blocking or prompting based on predefined rules about file access, commands, and environments. Recorded sessions log every tool invocation with the associated guard decision, providing an audit trail of what Claude Code executed. The system is designed to run entirely on local hardware, with no external data transmission. Token counts are derived from Anthropic’s own measurements and adjusted to reflect actual billing, enabling users to optimize configurations based on real usage patterns.
Key features
- Token tax analysis of request payloads
- Tool schema inventory and usage tracking
- Pre-execution guard with structured policy evaluation
- Flight recorder for executed tool calls
- Local browser interface at 127.0.0.1:8787
- Configurable deny lists for tools and commands
- Project-root-based path containment checks
- Deterministic policy enforcement without model queries
Use cases
- Reducing token usage by removing unused tool schemas
- Enforcing safe execution policies for file operations
- Auditing tool calls in automated Claude Code sessions
Pros
- Local proxy with no external data transmission
- Real-time token and tool usage analysis
- Pre-execution guard policies for tool calls
- Audit trail of executed tools and decisions
- Open-source and MIT licensed
Cons
- Requires Node.js 18+
- Desktop-only, terminal-integrated workflow
- No cloud or multi-user features described
- Policy changes apply only to new sessions
Frequently asked questions about Iris
What does Iris do?
Iris acts as a local proxy between Claude Code and Anthropic’s API, intercepting requests to reveal token composition, tool schemas, and conversation history. It evaluates tool usage, allows disabling unused schemas to reduce token load, and enforces execution policies before tool calls run.
Who is Iris designed for?
Iris is designed for developers and teams using Claude Code who want to optimize token usage, control tool access, and audit tool invocations. It is particularly useful for projects where minimizing costs and enforcing execution policies are priorities.
How does Iris help reduce token costs?
Iris shows which tool schemas are present in requests and whether they are actually used, allowing users to disable unused schemas. It measures token counts based on Anthropic’s reported usage, enabling users to optimize configurations and reduce unnecessary token consumption.
Can Iris enforce restrictions on tool execution?
Yes, Iris includes a guard system that evaluates tool calls against predefined rules before execution. It can block, allow, or prompt for confirmation based on policies for file access, commands, and environments, ensuring controlled execution.
Does Iris transmit data externally?
No, Iris runs entirely on local hardware and does not transmit any data outside the user’s machine. All processing and logging occur locally, ensuring privacy and compliance with data policies.
How do I get started with Iris?
To get started, install Iris via npm using the command `npx @hammadulhaq/iris` and access the interface at `127.0.0.1:8787` in your browser while running Claude Code. The tool provides a guided setup to configure token optimization and execution policies.
