crm.cli

A headless, CLI-first CRM for developers who do sales. Contacts, deals, and pipeline in a single SQLite file — queryable from your terminal, composable with Unix tools, and mountable as a virtual filesystem so any tool that reads files (Claude Code, Codex, grep, jq, vim) has full CRM access without any integration.

No server. No Docker. No accounts. No GUI. Just npm install -g @dzhng/crm.cli and go.

Created by Duet — a cloud agent workspace with persistent AI. Set up crm.cli in your own private cloud computer and run it with Claude Code or Codex — no local setup required. Try Duet →

Why crm.cli

Existing CRMs are GUI-first tools built for sales teams. If you're a technical founder, indie hacker, or engineer running BD, you're probably managing contacts in a spreadsheet you grep through. crm.cli is built for you.

Your CRM is a filesystem. Mount it with crm mount ~/crm and every tool that reads files — AI agents, shell scripts, editors — gets full CRM access for free. No MCP servers, no API keys, no integration code. The filesystem is the universal API.

crm mount ~/crm
ls ~/crm/contacts/
cat ~/crm/contacts/jane-doe.json | jq .name
# Point Claude Code at ~/crm and ask it to research your pipeline

Deep data normalization. A CSV has zero setup cost. crm.cli justifies its existence with structured intelligence: E.164 phone normalization (look up by any format), website normalization, social handle extraction (paste a LinkedIn URL, it stores the handle), entity merge with reference relinking, and fuzzy duplicate detection. This is what a spreadsheet can never provide.

crm contact add --name "Jane Doe" \
  --email [email protected] \
  --phone "+1-212-555-1234" \
  --linkedin linkedin.com/in/janedoe \
  --company Acme
# Phone stored as E.164, LinkedIn URL → handle, company auto-linked

Pipe-friendly by default. Every command outputs structured data. Pipe to jq, grep, awk, or feed into scripts. --format json on everything.

crm deal list --stage qualified --format json | jq '.[] | .value'
crm find "that fintech CTO from London"
crm report pipeline
crm dupes --threshold 0.5

Install

npm install -g @dzhng/crm.cli
# or: bun install -g @dzhng/crm.cli

# Or install the compiled binary
curl -fsSL https://raw.githubusercontent.com/dzhng/crm.cli/main/install.sh | sh

Storage

Everything lives in a single SQLite file. Default: ~/.crm/crm.db.

crm --db ./my-project.db contact list    # use a specific database
export CRM_DB=./team.db                  # or set via env var

No server. No Docker. No accounts. Back it up by copying the file.

Configuration

Config is loaded from crm.toml. Resolution order (first match wins):

1. --config <path> flag (explicit)
2. CRM_CONFIG env var
3. Walk up from CWD: ./crm.toml → ../crm.toml → ../../crm.toml → ... → /crm.toml
4. ~/.crm/config.toml (global fallback)

This means you can drop a crm.toml in your project root and it applies to everyone working in that directory — just like .gitignore or biome.jsonc.

# Project-scoped config
echo '[pipeline]
stages = ["discovery", "demo", "trial", "closed-won", "closed-lost"]' > ./crm.toml

# Global config (applies everywhere unless overridden)
mkdir -p ~/.crm
cat > ~/.crm/config.toml << 'EOF'
[database]
path = "~/.crm/crm.db"

[pipeline]
stages = ["lead", "qualified", "proposal", "negotiation", "closed-won", "closed-lost"]
won_stage = "closed-won"
lost_stage = "closed-lost"

[defaults]
format = "table"    # table | json | csv | tsv | ids

[phone]
default_country = "US"    # ISO 3166-1 alpha-2; for numbers without country code
display = "international" # international | national | e164

[hooks]
# pre-contact-add = "echo 'adding contact'"
# post-contact-add = "echo 'contact added'"
# pre-deal-stage-change = "echo 'stage changing'"

[mount]
default_path = "~/crm"          # where `crm mount` mounts by default
readonly = false                 # set true to prevent writes via FUSE
max_recent_activity = 10         # activities shown per entity in FUSE
search_limit = 20                # max results from search/find

EOF

Settings in a closer crm.toml override the global config. The --config flag overrides everything.

CLI Reference

Global Flags

Contacts

People you interact with.

crm contact add

crm contact add --name "Jane Doe" --email [email protected]
crm contact add --name "Jane Doe" --email [email protected] --email [email protected] --phone "+1-212-555-1234" --phone "+44-20-7946-0958" --company Acme --company "Acme Ventures" --tag hot-lead --tag enterprise
crm contact add --name "Jane Doe" --email [email protected] --linkedin janedoe --x janedoe --set title=CTO --set source=conference --set notes="Met at SaaStr"
crm contact add --name "Jane Doe" --linkedin https://linkedin.com/in/janedoe   # URL input also works — handle is extracted

Prints the created contact ID to stdout.

Social handles enforce uniqueness — no two contacts can share the same handle on a given platform. All of these input formats are accepted and normalized to the raw handle:

The same normalization applies to lookups, edits, and duplicate detection — crm contact show x.com/janedoe matches a contact stored with handle janedoe.

crm contact list

crm contact list
crm contact list --tag hot-lead
crm contact list --company Acme --format json
crm contact list --filter "title~=CTO AND source=conference"
crm contact list --limit 10 --offset 20

crm contact show <id-or-email-or-phone-or-handle>

crm contact show ct_01J8Z...
crm contact show [email protected]
crm contact show "+1-212-555-1234"
crm contact show janedoe                          # matches any social handle
crm contact show linkedin.com/in/janedoe          # URL also works — extracts handle

Accepts ID, any email, any phone number, or any social handle (LinkedIn, X, Bluesky, Telegram). URLs are also accepted — the handle is extracted before lookup. Shows full contact details including linked companies, deals, activity history, tags, and custom fields.

crm contact edit <id-or-email-or-phone-or-handle>

crm contact edit [email protected] --name "Jane Smith"
crm contact edit ct_01J8Z... --add-email [email protected] --rm-email [email protected]
crm contact edit ct_01J8Z... --add-phone "+44-20-7946-0958" --rm-phone "+1-310-555-9876"
crm contact edit ct_01J8Z... --add-company "Acme Ventures" --rm-company "Old Corp"
crm contact edit ct_01J8Z... --set title=CEO --set source=referral
crm contact edit [email protected] --add-tag vip --rm-tag cold

crm contact rm <id-or-email-or-phone-or-handle>

crm contact rm [email protected]
crm contact rm "+1-212-555-1234" --force    # skip confirmation

Prompts for confirmation unless --force is passed. Removes the contact and unlinks from deals/companies (does not delete linked entities).

crm contact merge <id> <id>

crm contact merge ct_01J8Z... ct_02K9A...

Merges two contacts. Keeps the first, absorbs data from the second. Emails, phones, companies, tags, custom fields, and activity history are combined. Deals linked to the second contact are relinked to the first. The first contact's name and social handles take priority. Prints the surviving contact ID.

Companies

Organizations that contacts belong to.

crm company add

crm company add --name "Acme Corp" --website acme.com
crm company add --name "Acme Corp" --website acme.com/ventures --website acme.co.uk --phone "+1-212-555-1234" --phone "+44-20-7946-0958" --tag enterprise --set industry=SaaS --set size=50-200

crm company list

crm company list
crm company list --filter "industry=SaaS" --format json
crm company list --tag enterprise --sort name

crm company show <id-or-website-or-phone>

crm company show acme.com
crm company show co_01J8Z...
crm company show "+1-212-555-1234"

Accepts ID, any stored website, or any phone number. Shows company details plus all linked contacts and deals.

crm company edit <id-or-website-or-phone>

crm company edit acme.com --name "Acme Inc" --set industry=Fintech
crm company edit co_01J8Z... --add-website acme.co.uk --add-phone "+44-20-7946-0958"
crm company edit acme.com --rm-website old-acme.com --rm-phone "+1-415-555-0000"

crm company rm <id-or-website-or-phone>

Prompts for confirmation unless --force is passed. Unlinks contacts and deals but does not delete them.

crm company merge <id> <id>

crm company merge co_01J8Z... co_02K9A...

Merges two companies. Keeps the first, absorbs data from the second. Websites, phones, tags, and custom fields are combined. All contacts and deals linked to the second company are relinked to the first. The first company's name takes priority. Prints the surviving company ID.

Deals

Pipeline tracking for opportunities.

crm deal add

crm deal add --title "Acme Enterprise" --value 50000
crm deal add --title "Acme Enterprise" --value 50000 --stage qualified --contact [email protected] --company acme.com --expected-close 2026-06-01 --probability 60 --tag q2

crm deal list

crm deal list
crm deal list --stage qualified
crm deal list --stage qualified --sort value --reverse
crm deal list --contact [email protected]
crm deal list --company acme.com
crm deal list --min-value 10000 --max-value 100000

crm deal show <id>

Shows full deal details including stage history (timestamps of every stage transition).

crm deal edit <id>

Same pattern as other entities. --stage is NOT used here — use crm deal move for stage changes (so transitions are tracked properly).

crm deal move <id> --stage <stage>

crm deal move dl_01J8Z... --stage negotiation
crm deal move dl_01J8Z... --stage closed-won --note "Signed annual contract"
crm deal move dl_01J8Z... --stage closed-lost --note "Budget cut"

Records the stage transition with a timestamp by creating a stage-change activity entry. This activity includes the old stage, new stage, and timestamp. Stage history is reconstructed from these activity entries — crm deal show includes a stage_history array with {stage, at} pairs. --note attaches a note to the transition (shown in crm report won and crm report lost).

Moving a deal to its current stage is rejected with an error.

crm deal rm <id>

Prompts for confirmation unless --force is passed.

crm pipeline

crm pipeline
crm pipeline --format json

Visual pipeline summary showing count, total value, and weighted value per stage.

Pipeline
────────────────────────────────────────────────────────
 lead          ████████████  12   $45,000    wt: $4,500
 qualified     ████████      8    $120,000   wt: $48,000
 proposal      ███           3    $85,000    wt: $42,500
 negotiation   ██            2    $60,000    wt: $48,000
────────────────────────────────────────────────────────
 Total: 25 deals | $310,000 pipeline | $143,000 weighted

Auto-Create Behavior

When linking entities via --contact or --company flags, crm.cli auto-creates stubs if the referenced entity doesn't exist. This applies consistently across all commands:

Auto-created contacts use the email local part as the name (e.g. [email protected] → name jane). Auto-created companies use the provided name directly. All stubs can be enriched later with crm contact edit or crm company edit.

--deal flags always require an existing deal — deals are not auto-created.

Activities

Interaction log attached to contacts, companies, or deals.

crm log <type> <body> [flags]

crm log note "Had a great intro call" --contact [email protected]
crm log call "Demo scheduled for Friday" --contact [email protected] --set duration=15m
crm log meeting "Went through pricing" --contact [email protected] --company Acme
crm log email "Sent proposal PDF" --contact [email protected] --deal dl_abc123
crm log note "General observation"  # standalone, no entity link

crm activity list

crm activity list
crm activity list --contact [email protected]
crm activity list --company acme.com
crm activity list --deal dl_01J8Z...
crm activity list --type call --since 2026-01-01
crm activity list --limit 20

Tags

Flat labels across all entity types.

crm tag <entity-ref> <tags...>

crm tag [email protected] hot-lead enterprise
crm tag acme.com target-account
crm tag dl_01J8Z... q2 high-priority

crm untag <entity-ref> <tags...>

crm untag [email protected] cold

crm tag list

crm tag list                    # all tags with counts
crm tag list --type contact     # tags used on contacts only

Search

crm search <query>

Exact keyword search using SQLite FTS5. Searches across all entity types and activity notes.

crm search "acme"
crm search "acme" --type contact
crm search "acme" --type contact,company
crm search "enterprise plan" --format json

crm dupes

Find likely duplicate entities using fuzzy matching on string fields. This is a review command for agents and operators — it suggests suspicious pairs, but does not merge anything.

crm dupes
crm dupes --type contact
crm dupes --type company --format json
crm dupes --limit 20

Typical signals:

• similar contact names with different emails
• similar company names with different websites
• same contact name + same company name
• similar social handles across contacts

Output includes both candidate entities plus the reasons they were flagged. Exact duplicates already prevented by uniqueness constraints (email, phone, social handles) should not appear here.

crm find <query>

Full-text keyword search across all entities using SQLite FTS5.

crm find "fintech"
crm find "enterprise SaaS"
crm find "jane"

Uses SQLite's built-in FTS5 full-text search — no API keys, no network calls, no external dependencies. The search index is automatically maintained as you add/edit entities.

crm index

crm index rebuild     # rebuild the full search index
crm index status      # show index stats (row count, last updated, model info)

The search index updates automatically on every write operation. Manual rebuild is only needed after config changes or corruption.

Reports

crm report pipeline

Pipeline summary with deal counts, values, and weighted values per stage.

crm report pipeline
crm report pipeline --format json
crm report pipeline --format markdown > pipeline.md

crm report activity

Activity volume over a time period.

crm report activity                       # last 30 days
crm report activity --period 7d           # last 7 days
crm report activity --period 90d --by type    # grouped by activity type
crm report activity --by contact          # grouped by contact

crm report stale

Contacts and deals with no recent activity.

crm report stale                    # no activity in 30 days (default)
crm report stale --days 7           # no activity in 7 days
crm report stale --type deal        # only stale deals
crm report stale --type contact     # only stale contacts

crm report conversion

Stage-to-stage conversion rates.

crm report conversion
crm report conversion --since 2026-01-01

Output:

Stage Conversion Rates
──────────────────────────────────────
 lead → qualified        42%  (50/120)
 qualified → proposal    65%  (33/50)
 proposal → negotiation  72%  (24/33)
 negotiation → won       58%  (14/24)
──────────────────────────────────────
 Overall: 12% lead-to-close

crm report velocity

Average time deals spend in each stage.

crm report velocity
crm report velocity --won-only     # only count deals that were won

crm report forecast

Weighted pipeline forecast by expected close date.

crm report forecast
crm report forecast --period Q2-2026
crm report forecast --period 2026-06    # single month

crm report won / crm report lost

Summary of closed deals.

crm report won --period 30d
crm report lost --period 30d

Both reports include a notes field with any notes attached via crm deal move --note.

Import / Export

crm import <entity-type> <file>

crm import contacts leads.csv
crm import contacts leads.json
crm import companies companies.csv
crm import deals deals.csv

CSV files expect headers matching core field names (name, email, phone, company, tags). Any unrecognized column headers are imported as custom fields. Tags are comma-separated within the field.

JSON files expect an array of objects with the same field names.

crm export <entity-type>

crm export contacts --format csv > contacts.csv
crm export deals --format json > deals.json
crm export all --format json > full-backup.json

crm export all exports the entire database (contacts, companies, deals, activities, tags) as a single JSON structure.

Bulk Operations

All list commands output IDs when using --format ids, enabling pipe-based bulk operations:

# Tag all contacts from Acme as enterprise
crm contact list --company Acme --format ids | xargs -I{} crm tag {} enterprise

# Move all qualified deals over $50k to proposal
crm deal list --stage qualified --min-value 50000 --format ids | xargs -I{} crm deal move {} --stage proposal

# Import from stdin
cat leads.json | crm import contacts -

# Chain with other tools
crm contact list --format json | jq '.[] | select(.tags | contains(["hot-lead"]))' | ...

Hooks

Shell commands triggered on mutations. Configured in ~/.crm/config.toml:

[hooks]
post-contact-add = "~/.crm/hooks/notify-slack.sh"
post-deal-stage-change = "~/.crm/hooks/deal-moved.sh"
pre-contact-rm = "~/.crm/hooks/confirm-delete.sh"

Hooks receive the entity data as JSON via stdin. Pre-hooks can abort the operation by exiting non-zero.

Available hooks:

• pre-* / post-* for: contact-add, contact-edit, contact-rm, company-add, company-edit, company-rm, deal-add, deal-edit, deal-rm, deal-stage-change, activity-add

Custom Fields

All entities (contacts, companies, deals, activities) support arbitrary custom fields via --set key=value. Custom fields store any JSON value — strings, numbers, booleans, arrays, objects.

# String value (default when using --set)
crm contact add --name "Jane" --set title=CTO --set source=conference

# JSON values (prefix with json: for non-string types)
crm contact edit [email protected] --set "json:score=85" --set "json:verified=true"
crm company edit acme.com --set "json:offices=[\"NYC\",\"London\"]"

# Remove a custom field
crm contact edit [email protected] --unset title

# Filter on custom fields (same syntax as core fields)
crm contact list --filter "title~=CTO"
crm company list --filter "industry=SaaS"

Only a small set of fields are hard-coded per entity — everything else lives in custom_fields:

Hard-coded fields drive business logic (entity resolution, pipeline math, relationships, reports). Custom fields are metadata — flexible, filterable, but no special behavior.

Phone Normalization

All phone numbers are normalized to E.164 format on input using libphonenumber-js. This ensures consistent storage, deduplication, and lookup regardless of how numbers are entered.

Storage: Always E.164 (+12125551234).

Display: Configurable via phone.display in crm.toml:

Input: Any reasonable format is accepted and normalized:

crm contact add --name "Jane" --phone "+1-212-555-1234"     # international with dashes
crm contact add --name "Jane" --phone "(212) 555-1234"      # national (uses default_country)
crm contact add --name "Jane" --phone "2125551234"           # digits only (uses default_country)
crm contact add --name "Jane" --phone "+12125551234"         # E.164
# All four store the same E.164 value: +12125551234

Lookup: Any format resolves to the same entity:

crm contact show "+12125551234"       # E.164
crm contact show "(212) 555-1234"     # national
crm contact show "212-555-1234"       # partial
# All three find the same contact

Dedup: All formats normalize to E.164, so duplicate detection works across formats:

crm contact add --name "Jane" --phone "+1-212-555-1234"
crm contact add --name "Bob" --phone "(212) 555-1234"    # fails: duplicate phone

Validation: Invalid numbers are rejected:

crm contact add --name "Jane" --phone "not-a-number"     # error: invalid phone number
crm contact add --name "Jane" --phone "123"               # error: too short

Country code: Numbers without a country code use phone.default_country from config (ISO 3166-1 alpha-2). If unset, a country code is required.

# crm.toml
[phone]
default_country = "US"
display = "international"

FUSE: _by-phone/ symlinks use E.164 filenames (+12125551234.json), so agents can look up phones in any format by normalizing first, or browse the directory for the canonical E.164 listing.

Website Normalization

Company websites are stored as full URLs.

Input is normalized for consistent storage and lookup:

• strip protocol if provided
• strip www. prefix
• lowercase the host
• preserve the path
• drop the trailing slash only when there is no path

crm company add --name "Acme" --website "https://www.Acme.com/labs"
# Stored as: acme.com/labs

This means path-based companies can remain distinct:

crm company add --name "Globex Research" --website "globex.com/research"
crm company add --name "Globex Consulting" --website "globex.com/consulting"

Dedup: exact match after normalization is a duplicate. The same normalized website cannot belong to two different companies:

crm company add --name "Acme Corp" --website "acme.com/labs"
crm company add --name "Acme Inc" --website "www.acme.com/labs"    # fails: duplicate website

Different paths are distinct:

crm company add --name "Globex Research" --website "globex.com/research"
crm company add --name "Globex Consulting" --website "globex.com/consulting"  # allowed — different path

Subdomains are distinct:

crm company add --name "Acme US" --website "us.acme.com"
crm company add --name "Acme EU" --website "eu.acme.com"            # allowed — different subdomain

Lookup: equivalent input formats resolve to the same normalized website:

crm company show "https://www.acme.com/labs"
crm company show "ACME.COM/labs"

FUSE: _by-website/ symlinks use the normalized website value, encoded as a filename-safe path.

Filtering

The --filter flag accepts expressions:

field=value                    # exact match
field!=value                   # not equal
field~=value                   # contains (case-insensitive)
field>value / field<value      # comparison (numeric/date fields)
expr AND expr                  # both conditions
expr OR expr                   # either condition

Custom fields are addressed by name — no prefix needed:

crm contact list --filter "title~=CTO AND company=Acme"
crm deal list --filter "value>10000 AND stage!=closed-lost"
crm contact list --filter "source=inbound OR source=referral"
crm company list --filter "industry=SaaS AND size~=50"

Output Formats

Every command that produces output supports --format:

Set a default via CRM_FORMAT env var or defaults.format in config.

IDs

All entities use prefixed ULID-based IDs:

Commands that accept an entity reference also accept email, phone, or social handle (contacts), or website URL/host or phone (companies) as shortcuts.

Virtual Filesystem (FUSE)

Mount the CRM as a read/write filesystem. AI agents (or humans) can explore CRM data using standard file operations — ls, cat, find, grep.

Mount

crm mount ~/crm                         # mount with default DB
crm mount ~/crm --db ./team.db          # mount a specific database
crm mount ~/crm --readonly              # read-only mode (no writes)
crm unmount ~/crm                       # unmount

The mount point stays live — changes made via the CLI or filesystem are reflected immediately in both directions.

Filesystem Layout

~/crm/
├── llm.txt                                # agent instructions (read this first)
├── contacts/
│   ├── ct_01J8Z...jane-doe.json           # full contact as JSON
│   ├── ct_02K9A...john-smith.json
│   ├── _by-email/                         # symlinks for email lookup
│   │   ├── [email protected] → ../ct_01J8Z...jane-doe.json
│   │   └── [email protected] → ../ct_02K9A...john-smith.json
│   ├── _by-phone/                         # symlinks for phone lookup (E.164 filenames)
│   │   └── +12125551234.json → ../ct_01J8Z...jane-doe.json
│   ├── _by-linkedin/                      # symlinks for LinkedIn handle lookup
│   │   └── janedoe.json → ../ct_01J8Z...jane-doe.json
│   ├── _by-x/                             # symlinks for X handle lookup
│   │   └── janedoe.json → ../ct_01J8Z...jane-doe.json
│   ├── _by-bluesky/                       # symlinks for Bluesky handle lookup
│   ├── _by-telegram/                      # symlinks for Telegram handle lookup
│   ├── _by-company/                       # symlinks grouped by company (contact appears under each)
│   │   └── acme-corp/
│   │       └── ct_01J8Z...jane-doe.json → ../../ct_01J8Z...jane-doe.json
│   └── _by-tag/                           # symlinks grouped by tag
│       ├── hot-lead/
│       │   └── ct_01J8Z...jane-doe.json → ../../ct_01J8Z...jane-doe.json
│       └── enterprise/
│           └── ct_01J8Z...jane-doe.json → ../../ct_01J8Z...jane-doe.json
├── companies/
│   ├── co_01J8Z...acme-corp.json
│   ├── _by-website/
│   │   ├── acme.com.json → ../co_01J8Z...acme-corp.json
│   │   └── acme.co.uk.json → ../co_01J8Z...acme-corp.json
│   ├── _by-phone/                         # E.164 filenames
│   │   └── +12125551234.json → ../co_01J8Z...acme-corp.json
│   └── _by-tag/
│       └── enterprise/
│           └── co_01J8Z...acme-corp.json → ../../co_01J8Z...acme-corp.json
├── deals/
│   ├── dl_01J8Z...acme-enterprise.json
│   ├── _by-stage/
│   │   ├── lead/
│   │   ├── qualified/
│   │   ├── proposal/
│   │   ├── negotiation/
│   │   ├── closed-won/
│   │   └── closed-lost/
│   ├── _by-company/
│   │   └── acme-corp/
│   │       └── dl_01J8Z...acme-enterprise.json → ../../dl_01J8Z...acme-enterprise.json
│   └── _by-tag/
│       └── q2/
├── activities/
│   ├── ac_01J8Z...note-2026-04-01.json
│   ├── _by-contact/
│   │   └── ct_01J8Z...jane-doe/
│   │       └── ac_01J8Z...note-2026-04-01.json → ../../ac_01J8Z...note-2026-04-01.json
│   ├── _by-company/
│   ├── _by-deal/
│   └── _by-type/
│       ├── note/
│       ├── call/
│       ├── meeting/
│       └── email/
├── pipeline.json                          # pipeline summary (read-only)
├── reports/
│   ├── pipeline.json                      # same as above
│   ├── stale.json                         # stale contacts/deals
│   ├── forecast.json                      # weighted forecast
│   ├── conversion.json                    # stage conversion rates
│   ├── velocity.json                      # time per stage
│   ├── won.json                           # closed-won summary
│   └── lost.json                          # closed-lost summary
├── tags.json                              # all tags with counts
└── search/                                # query by reading a "file"
    └── <query>.json                       # cat ~/crm/search/"acme CTO".json

File Format

Each entity file is a self-contained JSON document with linked data inlined:

// ~/crm/contacts/ct_01J8Z...jane-doe.json
{
  "id": "ct_01J8ZVXB3K...",
  "name": "Jane Doe",
  "emails": ["[email protected]", "[email protected]"],
  "phones": ["+12125551234"],
  "companies": [{ "id": "co_01J8Z...", "name": "Acme Corp" }],
  "linkedin": "janedoe",
  "x": "janedoe",
  "bluesky": null,
  "telegram": null,
  "tags": ["hot-lead", "enterprise"],
  "custom_fields": {
    "title": "CTO",
    "source": "conference"
  },
  "deals": [
    { "id": "dl_01J8Z...", "title": "Acme Enterprise", "stage": "qualified", "value": 50000 }
  ],
  "recent_activity": [
    {
      "id": "ac_01J8Z...",
      "type": "note",
      "note": "Had a great intro call",
      "created_at": "2026-04-01T10:30:00Z"
    }
  ],
  "created_at": "2026-03-15T09:00:00Z",
  "updated_at": "2026-04-01T10:30:00Z"
}

Read Operations

# Browse contacts
ls ~/crm/contacts/
cat ~/crm/contacts/ct_01J8Z...jane-doe.json

# Look up by email or phone
cat ~/crm/contacts/_by-email/[email protected]
cat ~/crm/contacts/_by-phone/+12125551234.json

# List all deals in a pipeline stage
ls ~/crm/deals/_by-stage/qualified/

# Get all contacts at a company
ls ~/crm/contacts/_by-company/acme-corp/

# Get all activities for a contact
ls ~/crm/activities/_by-contact/ct_01J8Z...jane-doe/

# Read pipeline report
cat ~/crm/pipeline.json

# Read stale report
cat ~/crm/reports/stale.json

# Search (read a virtual file whose name is the query)
cat ~/crm/search/"fintech CTO".json

# Use standard tools
grep -r "enterprise" ~/crm/contacts/
find ~/crm/deals/_by-stage/lead/ -name "*.json" | xargs jq '.value'

Write Operations

Writes are full-document replacements — the JSON you write becomes the complete state of the entity. This matches how files work: read → modify → write back.

Validation is strict. Every write is validated before being applied:

Errors are returned as the write syscall error, so agents get immediate feedback and can self-correct.

# Create a new contact (write any filename — ID is auto-assigned)
echo '{"name": "Bob Smith", "emails": ["[email protected]"]}' > ~/crm/contacts/new.json

# Update an existing contact — FULL document write (read, modify, write back)
jq '.custom_fields.title = "VP Engineering"' ~/crm/contacts/ct_01J8Z...jane-doe.json | \
  sponge ~/crm/contacts/ct_01J8Z...jane-doe.json

# Delete a contact
rm ~/crm/contacts/ct_01J8Z...jane-doe.json

# Move a deal stage (update the stage field in the full document)
jq '.stage = "closed-won"' ~/crm/deals/dl_01J8Z...acme-enterprise.json | \
  sponge ~/crm/deals/dl_01J8Z...acme-enterprise.json

# Log an activity (write to activities/)
echo '{"type": "note", "entity_ref": "[email protected]", "note": "Follow up on proposal"}' \
  > ~/crm/activities/new.json

# Error cases — all rejected with EINVAL
echo 'not json'                  > ~/crm/contacts/new.json    # parse error
echo '{"bogus": 1}'             > ~/crm/contacts/new.json    # unknown field
echo '{"emails": ["[email protected]"]}'  > ~/crm/contacts/new.json    # missing "name"
echo '{"name": "X", "emails": "wrong"}' > ~/crm/contacts/new.json  # type mismatch

AI Agent Usage

The filesystem is designed for agents that navigate via ls, cat, and find. A llm.txt file at the root describes the filesystem structure and usage — agents should read this first. An agent exploring CRM data might:

Agent: Let me understand this filesystem.
> cat ~/crm/llm.txt

Agent: Let me check the pipeline.
> cat ~/crm/pipeline.json

Agent: Who are the contacts at Acme?
> ls ~/crm/contacts/_by-company/acme-corp/
> cat ~/crm/contacts/_by-company/acme-corp/ct_01J8Z...jane-doe.json

Agent: What deals are stale?
> cat ~/crm/reports/stale.json

Agent: Let me search for that fintech CTO.
> cat ~/crm/search/"fintech CTO london".json

Agent: I'll update her title.
> jq '.custom_fields.title = "CEO"' ~/crm/contacts/ct_01J8Z...jane-doe.json > /tmp/update.json
> mv /tmp/update.json ~/crm/contacts/ct_01J8Z...jane-doe.json

Configuration

FUSE settings in crm.toml:

[mount]
default_path = "~/crm"          # default mount point for `crm mount`
readonly = false                 # default to read-write
max_recent_activity = 10        # how many activities to inline in entity files
search_limit = 20               # max results for search/ virtual files

How Mount Works

crm mount starts two background processes:

1. Daemon (crm __daemon) — a Unix socket server that handles all business logic (reads, writes, search, validation, normalization). The crm binary spawns itself with a hidden __daemon subcommand, so this works identically whether installed via npm, compiled binary, or running from source during development.

2. Filesystem bridge — connects the OS filesystem layer to the daemon:

   • macOS: An NFS v3 server (Rust binary, auto-compiled on first mount via Cargo) that translates NFS operations to daemon socket calls. Mounted via mount_nfs.
   • Linux: A FUSE helper (C binary, auto-compiled on first mount via gcc) that translates FUSE syscalls to daemon socket calls.

Platform Notes

• Linux: Works out of the box (FUSE is in the kernel). Requires libfuse3-dev and gcc.
• macOS: Requires Rust toolchain for the NFS server (curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh). No FUSE or kernel extensions needed.
• Windows: Not supported (use WSL).
• Containers/sandboxes: FUSE requires --privileged or --device /dev/fuse. If unavailable, use the CLI with --format json instead.

Stack

• Runtime: Bun
• Language: TypeScript (strict mode)
• Database: SQLite via libSQL + Drizzle ORM
• Phone normalization: libphonenumber-js (E.164)
• Website normalization: normalize-url
• Linting: Biome via Ultracite
• Testing: bun test (functional tests at the CLI level)

Distribution

Package Manager

npm install -g @dzhng/crm.cli
# or: bun install -g @dzhng/crm.cli

Compiled Binary

Pre-built binaries are available on GitHub Releases, compiled via bun build --compile for:

• Linux x64
• Linux arm64
• macOS x64 (Intel)
• macOS arm64 (Apple Silicon)

# Download and install the latest release
curl -fsSL https://raw.githubusercontent.com/dzhng/crm.cli/main/install.sh | sh

Install Script

The install script (install.sh) handles:

1. Detecting your platform (Linux/macOS, x64/arm64)
2. Downloading the correct binary from GitHub Releases
3. Installing it to ~/.local/bin (or /usr/local/bin with sudo)
4. Installing FUSE dependencies:
   • Linux: libfuse3-dev via apt/yum/pacman
   • macOS: macfuse via Homebrew

curl -fsSL https://raw.githubusercontent.com/dzhng/crm.cli/main/install.sh | sh

CI/CD

GitHub Actions pipeline:

• On push to main: Run all tests via bun test
• On tag (v*): Build binaries for all platforms, publish to npm, create GitHub Release with assets

Created by Duet

crm.cli is created by Duet — a cloud agent workspace where every user gets a private cloud computer with a persistent, always-on AI agent. Set up crm.cli in your own Duet workspace and run it with Claude Code or Codex in the cloud — no local setup required.

Try Duet →

License

MIT