intaris 0.3.1

Guardrails service for AI agents — safety evaluation, audit, and approval workflows

intaris

Guardrails service for AI agents. Intaris sits between your AI agent and its tools, evaluating every tool call for safety and alignment before allowing execution. Works with OpenCode, Claude Code, OpenClaw, and any MCP-compatible client.

Default-deny. Every tool call is classified and evaluated. Read-only operations are fast-pathed; everything else goes through LLM safety evaluation. Unknown tools are never auto-approved.

Real-time. Sub-second evaluation with a priority-ordered decision matrix. Read-only calls resolve in under 1ms. LLM evaluations complete within the 5-second circuit breaker. WebSocket streaming for live monitoring.

Self-hosted. Single Python process, SQLite or PostgreSQL storage, no external dependencies beyond an LLM API key. Your code and audit trail stay under your control.

Part of the Cognara platform (Cognis controller, Intaris guardrails, Mnemory memory).

Features

• Default-deny classifier -- Explicit read-only allowlist with critical pattern detection. Everything not allowlisted goes through LLM evaluation.
• LLM safety evaluation -- OpenAI-compatible structured output for alignment checking, risk assessment, and decision reasoning.
• Priority-ordered decision matrix -- Critical risk auto-denies, aligned low/medium approves, high risk and misalignment escalate for human review.
• Session management -- Hierarchical parent/child sessions with intention tracking, lifecycle states, and idle sweep.
• Intention tracking -- User-driven intention model with IntentionBarrier for real-time updates and AlignmentBarrier for parent/child enforcement.
• MCP proxy -- Sits between clients and upstream MCP servers, evaluating every tool call with per-tool preference overrides.
• Audit trail -- Every evaluation is logged with decision, reasoning, risk level, classification, latency, and redacted arguments.
• Secret redaction -- API keys, passwords, tokens, and connection strings are automatically redacted before audit storage.
• Filesystem path protection -- Working directory enforcement with approved path prefix learning from LLM approvals.
• Session recording -- Full-fidelity event logs with live tailing, playback, and chunked ndjson storage (filesystem or S3).
• Behavioral analysis -- Three-layer system: per-call data collection, session summaries, and cross-session behavioral profiling.
• Management UI -- Built-in web dashboard with session tree view, audit log, approval queue, MCP server management, and real-time charts.
• Judge auto-resolution -- Escalated tool calls can be automatically reviewed by a more capable LLM (gpt-5.4), reducing human intervention while maintaining safety. Three modes: disabled, auto, advisory.
• Webhook callbacks -- HMAC-signed escalation notifications for external approval systems.
• Notification channels -- Per-user push notifications (Pushover, Slack, webhook) with one-click approve/deny action links.
• Rate limiting -- Per-session sliding window rate limiter to prevent runaway agents.

Quick Start

Intaris needs an OpenAI-compatible API key for safety evaluation. It picks up LLM_API_KEY from your environment automatically.

LLM_API_KEY=sk-your-key uvx intaris

That's it. Intaris starts on http://localhost:8060, management UI at http://localhost:8060/ui.

Now integrate with your agent. We already ship extensions for some clients. For example for OpenCode, install the plugin:

export INTARIS_URL=http://localhost:8060
cp integrations/opencode/intaris.ts ~/.config/opencode/plugins/

Intaris can also serve as MCP proxy with audit trail and guardrails for tool calls. To use that, configure any MCP client to use intaris as a single MCP server:

{
  "mcpServers": {
    "intaris": {
      "type": "streamable-http",
      "url": "http://localhost:8060/mcp"
    }
  }
}

And add MCP servers via Intaris UI or config.

Intaris is also available via Docker, pip, or production setup. See the full quick start guide for more clients and options.

Screenshots

Dashboard -- evaluation metrics, decision distribution, performance stats, and activity timeline

Sessions -- hierarchical tree view with expandable session details and recent evaluations

Approvals -- pending escalations with reasoning, arguments, and one-click approve/deny

Analysis -- behavioral risk profile with per-agent risk indicators and trends

Analysis -- cross-session behavioral trend tracking over time

Sessions -- suspicious session detail with evaluation reasoning and risk assessment

Audit -- critical tool execution denied with detailed reasoning

See the Management UI docs for all tabs and features.

Supported Clients

Client	Integration	Setup Guide
OpenCode	Plugin (intaris.ts)	OpenCode Guide
Claude Code	Hooks (bash scripts)	Claude Code Guide
OpenClaw	Plugin (@fpytloun/openclaw-intaris)	OpenClaw Guide
Any MCP client	MCP proxy (/mcp endpoint)	MCP Proxy Guide

Plugin/Hooks give fine-grained control: custom error messages, fail-open/fail-closed behavior, session lifecycle management, and behavioral analysis. MCP proxy is zero-code configuration but with less UX control.

How It Works

Intercept. The client integration (plugin, hooks, or MCP proxy) captures every tool call before execution and sends it to Intaris for evaluation.

Classify. The classifier checks the tool against a priority chain: session policy denies, tool preference overrides, critical patterns, the read-only allowlist, and filesystem path policy. Read-only tools are auto-approved. Critical patterns are auto-denied.

Evaluate. Tool calls classified as WRITE go through LLM safety evaluation. The LLM assesses alignment with the session intention, risk level (low/medium/high/critical), and recommends a decision -- all within a 4-second timeout.

Decide. The decision matrix applies priority-ordered rules: critical risk always denies, aligned low/medium risk approves, high risk and misalignment escalate for human review. The decision, reasoning, and full context are recorded in the audit trail.

See the Architecture and Evaluation Pipeline docs for the full technical details.

Benchmark Results

Intaris catches 100% of critical threats (destructive commands, data exfiltration, RCE) with zero false positives. Across 41 benchmark scenarios including adversarial attacks, social engineering, and cross-session patterns, Intaris achieves 94% F1 with 100% precision -- it never blocks legitimate developer work.

Metric	Value
Precision	100%
F1 Score	93.7%
False Positive Rate	0.0%
Critical Misses	0
Avg Latency	1.1s

See the Benchmarking docs for methodology, scenario details, and how to run your own benchmarks.

Documentation

Document	Description
Quick Start	Get running in 5 minutes
Architecture	System design, layers, and key decisions
Evaluation Pipeline	Classification, LLM evaluation, and decision matrix
Configuration	Environment variable reference
REST API	Full API endpoint reference
MCP Proxy	MCP proxy setup, tool namespacing, and preferences
Management UI	Built-in web dashboard
Deployment	Production deployment guide
Development	Contributing, tests, and code conventions
OpenCode Integration	OpenCode plugin setup
Claude Code Integration	Claude Code hooks setup
OpenClaw Integration	OpenClaw extension setup
Benchmarking	Guardrails benchmark system

License

Business Source License 1.1 — see LICENSE for the full text.

The Licensed Work is (c) 2026 Filip Pytloun. You may use the Software for your own internal business operations free of charge. Commercial use (SaaS, managed services, or as a component of a commercial product) requires a separate license. On the Change Date (2030-03-15), the license converts to Apache License 2.0.

For alternative licensing arrangements, contact: [e-mail]