Human-first architecture
Traditional CRMs assume someone logs in, clicks tabs, types into fields, and hits save. Every interaction assumes eyes on a screen. AI agents need structured, deterministic API access, not a UI.
The CRM Built for
AI Agents, Not Humans
Headless, local-first, evidence-based. 29 REST endpoints. 27 MCP tools. Zero vendor lock-in. $0 to run forever.
terminal
REST Endpoints
29
MCP Tools
27
Forever
$0
License
Apache 2.0
Enterprise SaaS was built
for humans clicking buttons.
AI agents are a different kind of user, and the software they've inherited was never designed for them.
$150 to $330 per seat, per month
A 10-person team on Salesforce pays $18K to $40K a year in licensing before you add Einstein AI, Slack, and security review. That's the cost of architectural decisions made in the 1990s.
Your data lives on their servers
Salesforce controls your data residency, change control, and export terms. If you need to migrate, it's expensive and disruptive. You can't even query your own data with local tools.
Flat fields that lose context
CRMs store Deal Amount and Close Date, not why the deal is worth that amount or who confirmed it. When an agent updates a value, the old one vanishes. No receipts, no history, no reasoning.
Built from the ground up
for agents as the primary user.
Not a CRM with an API bolted on. An agent-first data layer with a clean interface for every client that needs it.
Intentionally Headless
No frontend ships with Open Tooling CRM. The data layer and agent tooling are the product. The interface is whatever you want: scaffold a UI in minutes or skip it entirely and interact through Claude.
Evidence-Based Memory
Every claim traces back through a structured chain: Artifacts → Observations → Briefs → Conflicts. When your agent says "deal is $250K," you can trace that to the exact email where the prospect confirmed it.
Local-First & Self-Hosted
SQLite-backed with zero cloud dependencies. Run it on a $5/month VPS. Query your own data directly with SQL. Move it to a new server in minutes. Full compliance with data residency requirements.
MCP-Native
27 tools purpose-built for Claude Desktop, Claude Code, and any MCP-compatible client. Claude understands your data structure natively, with no translation layer and no schema gymnastics.
Fully Configurable
Define your own entity types, wire up your own integrations, deploy for your specific workflow. A construction company can track permit status. A recruiter can track candidates. No vendor assumptions.
Apache 2.0 Licensed
Commercial use Modification Distribution Patent grant Private use Sublicense
$0 forever
You own it. You run it. You fork it. No enterprise license gates, no usage caps, no surprise pricing tiers. Open Tooling CRM is free in perpetuity, commercially usable with full source access.
Free forever Get Started on GithubEvery claim has receipts.
Traditional CRMs store fields. Open Tooling stores evidence: an immutable, traceable chain from raw data to actionable intelligence.
Artifacts
Raw, immutable evidence: emails, transcripts, documents. Content-hashed for integrity. timestamped and append-only. Nothing ever overwrites an artifact.
Observations
Typed claims extracted from artifacts. Each has a lifecycle: current, superseded, or retracted. Old observations are preserved, never deleted.
Briefs
Human-readable summaries derived from observations. Always regeneratable from evidence. Agents read briefs first, drill to raw evidence when needed.
Conflicts
Explicit records when observations disagree. Never silently resolved. Budget approved (Oct 15) vs Budget rejected (Oct 20), both live, flagged for review.
Why this matters when agents are making decisions
Most CRMs store Deal Amount: $250,000, a field that can be overwritten without a trace. When your agent updates it, the old value vanishes.
Open Tooling stores the evidence behind the number. The observation links to the artifact: the email from the prospect's CFO where she confirmed the budget. The old observation is superseded, not deleted. The full audit trail is always there.
Real sales conversations are messy. "We might close in Q4" is an observation with uncertainty attached, not a Close Date field the agent mistakes for a hard commitment.
When your data has evidence and provenance, your agents reason like humans. Without it, they hallucinate with confidence.
Read the full architectureClaude understands your CRM natively.
27 MCP tools, drop-in for Claude Desktop, Claude Code, and any MCP-compatible client. No plugin config, no schema docs, no translation layer — or install the one-click Cowork plugin from the Attri marketplace.
Browse Plugin Marketplace
27
MCP Tools
29
REST Endpoints
create_entity search_entities unlink_entities ingest_artifact supersede_observation get_brief full_text_search list_activities export_graph
update_entity archive_entity list_relationships add_observation retract_observation list_conflicts semantic_search bulk_import
get_entity link_entities traverse_graph list_observations create_brief resolve_conflict get_timeline export_entities
Three commands. Running in minutes.
No Kubernetes. No paid cloud tier. No enterprise sales call. Clone, configure, run.
One command from GitHub. The CRM module lives in /crm. The full monorepo includes the plugin marketplace and future modules.
npm install, copy the example env file. SQLite database initializes automatically, with no separate database setup and no cloud credentials.
Server starts at localhost:8787. Run npm run seed to load sample data and explore the evidence chain with real records.
Add the MCP server to your Claude Desktop config. Or install the Cowork plugin from the Attri marketplace: one click, zero config, Claude is ready to work with your CRM in natural language.
REPO · Attri-Inc/open-tooling
github.com /open-tooling
STATUS
PUBLIC $ git clone github.com/Attri-Inc/open-tooling.git
Cloning into 'open-tooling'...
remote: Counting objects: 2347, done.
✓ Cloned in 3.2s
BRANCH
main
STARS
2.3k
LICENSE
Apache 2.0
INSTALL · open-tooling/crm
SQLite (auto-init)
STATUS
READY $ cd open-tooling/crm
$ npm install
added 247 packages in 8s
$ cp .env.example .env
✓ Database initialized
DEPS
247
DB
SQLite
ENV
.env
SERVER · open-tooling/crm
localhost: 8787
STATUS
RUNNING $ npm run dev
✓ Server running at localhost:8787
✓ REST API mounted at /api/v1
✓ MCP Server mounted at /mcp
PORT
8787
REST
29 endpoints
MCP
27 tools
CLIENT · Claude Desktop
natural language ready
STATUS
CONNECTED // claude_desktop_config.json
"mcpServers": {
"crm": {
"url": "localhost:8787/mcp"
}
}
CLIENT
Desktop
TOOLS
27 ready
AUTH
Local
Open Tooling vs Salesforce
Not a fair fight on cost. A different kind of fight on architecture.
Feature
Salesforce
Open Tooling CRM
Pricing
$150 to $330 / seat / month
$0, self-hosted forever
Primary user
Humans using a UI
AI agents via MCP + REST
MCP support
Requires custom plugin build
27 tools, native
Data ownership
Salesforce servers, their terms
Your infrastructure, your SQL
Memory model
Flat fields, overwritten on save
Artifacts → Observations → Briefs → Conflicts
Audit trail
Field history (limited)
Full evidence chain, every claim sourced
Offline / on-prem
Cloud-only
SQLite, runs fully offline
Customization
Config clicks, paid add-ons
Code: any entity type, any domain
License
Proprietary, per-seat
Apache 2.0, open source forever
CRM is just the first module.
Open Tooling is an open-source family of AI-native SaaS tools: the same headless, evidence-based, agent-first architecture across your entire back office.
- CRMAvailable now
- HRComing soon
- FinanceComing soon
- ProjectsComing soon
- ProcurementComing soon
- MoreIn exploration
Your agents deserve
better infrastructure.
Free to run. Open to fork. Ready for the AI workflows you're building today.
- $0 forever
- Apache 2.0
- Self-hosted
Apache 2.0 licensed · Built by Attri AI · Contributions welcome