Show HN: Conduct, open-source guardrails for LLM and MCP tool calls

Two merchandise surfaces, 1 repo, 1 policy:

• Conduct Guard — the argumentation engine. Decides artifact / pass / audit / inject for each AI action before it executes, backed by signed configuration and a hash-chained audit log.
• Conduct Router — the LLM proxy. Point immoderate supplier SDK (Anthropic, OpenAI, Perplexity) astatine Router and each petition runs done Guard connected the measurement to the upstream provider.

Runtime firewalls for illustration Straiker and Lakera show you what an supplier did. Guard controls what an supplier can do — pinch cryptographic proof.

The three-pillar moat:

1. Signed configuration — each workspace signs its progressive argumentation set. Every Guard cheque verifies the signature earlier enforcing. A tampered battalion — pushed by anyone, astatine immoderate furniture — is rejected earlier it tin determine anything.
2. Hash-chained audit — each determination appends to a SHA-256 concatenation rooted astatine workspace genesis. Any missing aliases altered introduction breaks the concatenation and is caught connected one-click verification. Evidence you tin manus to an auditor.
3. Policy-first, not detection-first — rules determine earlier the action executes, pinch system reasons. Not anomaly discovery aft the fact.

New here? Start pinch Discovery mode: read-only visibility into each AI action your squad takes for 14 days. No argumentation to author, thing to instal upstream, nary cost. When you're fresh to enforce, beforehand a norm from what Discovery already saw.

→ conductai.ai/sign-up

• API connected http://localhost:8000 (Guard + Router unrecorded astatine /guard/* and /proxy/*)
• Canvas UI connected http://localhost:3000
• Redis worker + Postgres travel up successful the aforesaid stack

Point immoderate supplier SDK astatine Router:

Or wrap your CLI hooks pinch Guard:

Now each Claude Code, Cursor, Copilot, ChatGPT, aliases Codex convention connected that instrumentality is governed by the aforesaid progressive packs.

20+ compliance packs vessel retired of the box: OWASP, SOC 2 CC7.3, HIPAA §164.312, PCI DSS 4.0, EU AI Act Art. 15/16, NIST AI RMF, ISO 42001, and framework-specific packs for Python, Node, and Terraform.

22 pre-built playbooks: Issue → PR, codification review, incident response, prod deploy gate, CI/CD triage, information scanner triage, Slack digest, and more. Each is 1 YAML file; edit-and-run.

Guard checks occurrence astatine 3 chokepoints:

• CLI hook — each Claude Code / Cursor / Copilot / Codex instrumentality call.
• MCP layer — each MCP instrumentality invocation.
• Router — each LLM telephone by immoderate SDK.

One policy, 3 enforcement surfaces.

• Self-host pinch docker compose — the bid above. Runs everything locally.
• Self-host connected Kubernetes — deployment templates vessel successful issue #1149.
• Hosted — conductai.ai.tier includes Discovery; paid tiers unlock enforcement + Router + hash-chain verification API.

• SECURITY.md — vulnerability reporting policy, scope, coordinated disclosure, and safe harbor.
• Threat model — strategy context, spot boundaries, attacker goals, mitigations, and residual risks.
• Policy determination contract — guard_check determination semantics and fail-mode behavior.
• Audit log verification — independent prev_hash/entry_hash concatenation verification process and illustration script.
• API versioning — proxy/MCP compatibility, deprecation windows, and OpenAPI publication guidance.

Apache License 2.0 — the full repository, including the CLI, Guard, Router, Agent Booster, playbooks, and packs.

• Free for commercialized and non-commercial use, modification, and redistribution.
• Includes an definitive patent assistance from each contributors (Apache 2.0 §3).
• Trademark authorities are not granted; spot NOTICE — "Conduct", "Conduct AI", and "Conduct Guard" stay trademarks of Conduct AI.
• Redistribution must sphere the LICENSE and NOTICE files.

The hosted power level astatine conductai.ai (canvas UI, squad RBAC, marketplace, managed Guard) is simply a commercialized offering built connected apical of this repository.

For endeavor support, indemnification, aliases licensing questions, email [email protected].

We judge bug reports, docs fixes, caller playbooks, caller packs, tests, and code. Read CONTRIBUTING.md first.

• Everyone participating agrees to the Code of Conduct.
• Security vulnerabilities: don't unfastened a nationalist issue. See SECURITY.md.
• Anything else: GitHub Discussions aliases SUPPORT.md.

• Product: conductai.ai
• Guard landing: conductai.ai/guard
• Router landing: conductai.ai/router
• Docs: conductai.ai/docs
• Discussions: github.com/sseshachala/conductai/discussions
• Changelog: CHANGELOG.md + Releases
• Book a demo: cal.com/sudhi-seshachala-pks7pd

⭐ If Conduct saves your squad time, star it — it helps different teams find it.