๐Ÿ‡จ๐Ÿ‡ฆ The compliance-first claw variant

AI agents that pass
your compliance audit

CASL consent management. PIPEDA audit trails. Default-deny networking. Forked from NanoClaw. Built for Canadian enterprise. Open source.

# Clone, setup, done
$ gh repo fork aiforcanadians/northclaw --clone
$ cd northclaw && claude
# Then type: /setup
5
Security layers
<5K
Lines of code
3
Privacy laws built in
0%
Malicious skills
219
Tests passing
0
CVEs
Architecture

Six layers between your data and the internet

No other agent framework enforces Canadian privacy law. NorthClaw makes compliance automatic, not optional.

01๐Ÿ›ก๏ธ

CASL Consent Gate

Every outbound message passes through real-time consent verification. Commercial vs transactional classification, jurisdiction detection, automatic CASL field injection. Runs on the host, outside agent containers. A compromised agent cannot bypass it.

02๐Ÿ”’

Default-Deny Egress

Agents cannot reach the internet. Period. Docker --internal network blocks all outbound traffic. The credential proxy is the only path out. Even under prompt injection, your data stays put.

03๐Ÿ“ฆ

Container Isolation

Every agent runs in its own sealed container. Read-only root filesystem, seccomp syscall filtering, no-new-privileges flag. One agent cannot see another agent's data. OS-level walls, not app-level promises.

04๐Ÿ“‹

Tamper-Evident Audit

Every agent action logged with SHA-256 hash chain. Each entry references the previous. Tampering breaks the chain. Stored outside container reach. When regulators ask what your AI did, you can prove it.

05๐Ÿ

Canadian Data Residency

Persistent data stays on Canadian infrastructure. Consent database, audit logs, agent memory, all local. Inference routing through nearest endpoints with full sovereignty roadmap via TELUS AI Factory.

06โšก

Skills, Not Features

New capabilities via SKILL.md files. No PRs to main branch. The core stays lean. Pipeline briefings, meeting debriefs, proposal drafting, compliance exports. Add what you need, nothing you don't.

ClearClaw

The honest comparison

We built this. We're biased. We also show you the data.

NorthClawOurs
TS ยท <5K
Security
9.5
Compliance
10
Performance
8
Auditability
10
Sovereignty
9
Ecosystem
6
NemoClaw
Python ยท ~20K
Security
7
Compliance
2
Performance
5
Auditability
5
Sovereignty
7
Ecosystem
7
NanoClaw
TS ยท ~3.9K
Security
7
Compliance
1
Performance
8
Auditability
10
Sovereignty
3
Ecosystem
5
OpenClaw
TS ยท ~430K
Security
3
Compliance
1
Performance
4
Auditability
1
Sovereignty
2
Ecosystem
10
ZeroClaw
Rust ยท ~15K
Security
8.5
Compliance
1
Performance
10
Auditability
7
Sovereignty
4
Ecosystem
6
IronClaw
Rust ยท ~12K
Security
10
Compliance
1
Performance
7
Auditability
8
Sovereignty
6
Ecosystem
3

Scores reflect our assessment. OpenClaw has 13,729 skills, 20% flagged as malicious. ZeroClaw boots in 10ms on 8MB of RAM, but the real bottleneck for agent tasks is API latency, not local compute. IronClaw has WASM sandboxing that can prove agent behavior cryptographically. Nobody needs that for sending follow-up emails. NorthClaw competes on compliance and trust, not on startup milliseconds.

Security Model

What happens when an agent gets prompt-injected?

In OpenClaw: it exfiltrates your data. In NorthClaw: it can't reach the internet.

1
ContainerEnforced

Agent runs in sealed Docker container. Read-only filesystem. Seccomp syscall filtering. 120s hard timeout. Destroyed after use.

2
NetworkDefault deny

Docker --internal network. No internet access. Credential proxy is the only path out. Even DNS is blocked.

3
CredentialsIsolated

API keys never enter containers. OAuth tokens never enter containers. All external calls proxied through host.

4
AuditImmutable

Every action logged with SHA-256 hash chain. Append-only. Tamper detection via chain verification. Stored outside container reach.

5
ComplianceHost-level

CASL consent gate runs on host. PIPEDA logging on host. Law 25 decision logging on host. Compromised agent cannot bypass.

Get Started

Three commands. Five minutes.

Clone the repo, run setup, connect Slack. Claude Code handles everything else.

# 1. Fork and clone
$ gh repo fork aiforcanadians/northclaw --clone
$ cd northclaw
# 2. Setup (Claude Code handles deps, Docker, auth)
$ claude
> /setup
# 3. Connect Slack
> /add-slack
Documentation

Built-in skills for real work

/pipeline

Daily briefing. Overdue follow-ups, today's meetings, stale contacts.

/debrief

Post-meeting processing. Decisions, commitments, intelligence extracted.

/discover

Pre-meeting intel. Web research, past context, conversation approach.

/proposal

Draft proposals. Value calculation before pricing. Verify pass included.

/value-report

Monthly client value reports. 5-layer model with evidence.

/egress add

Manage network allowlist. One command to allow a domain.

/consent-export

Export consent records for regulatory requests. JSON, CSV, or text.

/weekly

Monday review. Revenue, pipeline, decisions, content opportunities.

/new-skill

Create new skills from a template. Extend without touching core.

Built by

AI for Canadians

NorthClaw is built by AIFC, founded by Patrick Farrar. AI Entrepreneur in Residence at DMZ, Canada's #1 ranked university incubator. 500+ Canadian ecosystem partnerships. $18M+ in funding enabled. 90+ web applications. 30+ custom AI tools delivered.

DMZยทGenome AtlanticยทTELUSยทCommunitechยทPropel ICT