Skip to main content

Internal admin/marketing CLI — discovery, enrichment, outreach, and CRM over one interface.

Project description

dogfu

Internal admin CLI + SDK for Agent Berlin staff (admin_users). One interface over LinkedIn, X, Google, ChatGPT, SEO (the dogfu data API) and the Close CRM (via the admin CRM proxy). The sibling of backend/sdk-python — same transport posture (a single Bearer token against the Berlin backend, retry with backoff, a typed exception hierarchy, config-file session), scoped to the internal admin surface instead of a customer project.

Auth

dogfu authenticates with a single session token (a short-lived JWT, 24h), minted by exchanging a one-time OTP from the dogfu MCP's get_setup_instructions:

dogfu configure --otp <OTP> --title "<what this session is for>"

This POSTs {otp, title} to /dogfucli/exchange-otp and saves the returned token to ~/.dogfu/config.json. Subsequent commands resolve the token from DOGFU_TOKEN first, then that file. There is no refresh endpoint — on an auth failure, run get_setup_instructions again for a fresh OTP and re-configure.

dogfu is not project-scoped (no --domain); access is platform-wide for staff.

Endpoints

Everything goes through the Berlin backend (https://backend.agentberlin.ai by default; override with --base-url / DOGFU_BASE_URL):

Group Backend route
linkedin / x / google / chatgpt / seo / apollo / report /api/v1/admin/dogfu/*
crm (leads, contacts, tasks, notes, opportunities, pipelines) /api/v1/admin/crm/* — allowlisted proxy to Close CRM

The CRM proxy holds the per-admin Close key server-side (configured from the admin Console), so dogfu never sees a Close credential — it forwards Close sub-paths under the dogfu session token.

Usage

dogfu --help                  # list groups
dogfu <group> --help          # list a group's commands
dogfu <group> <cmd> --help    # flags + the Output: data shape

dogfu seo domain-overview --target acme.com
dogfu crm lead create --name "Acme" --url https://acme.com

Output is canonical JSON by default (-f table for humans, -o FILE to write).

Outreach cadence

A touch is an attempt to reach a lead — touch 0 is the reach-out, 1..N are follow-ups, on whatever channel the rep used. The sequence is unbounded and only ends on reply or stop. State lives in four system-managed lead custom fields (touch_stage, last_touched, next_touch_due, touch_channel), created once by a Close admin; only the crm touch verbs write them.

dogfu crm worklist                  # the work queue: every open task due today, classified
dogfu crm touch record <lead_id> -c linkedin   # log a touch, roll the reminder forward
dogfu crm touch reply <lead_id>     # prospect answered: end the chase
dogfu crm touch stop  <lead_id>     # give up / nurture: end the chase
dogfu crm reconcile                 # audit the task invariants (--apply to repair)

Every lead in the active sequence carries exactly one open cadence task — created when the lead enters the reach-out status (crm lead create/update -s Qualified), rolled forward by record, and closed on reply/stop.

The work queue is the task list. crm worklist reads the actual open Close tasks due today or earlier — the BDR's real inbox — and classifies each row (reach-out / follow-up / deal / ad-hoc), enriching it with the lead and contact context to act. It replaces the old touch due / opportunity due commands, so crm worklist and Close's own Tasks/Inbox are literally the same list — nothing due is ever invisible, and there's no state-vs-task reconstruction to drift.

crm reconcile is the inverse — the audit: everything whose state says it needs action but has no matching task (a reach-out/cadence task missing, a dropped-ball opportunity, a stray/duplicate/terminal-status task, a stalled deal). With --apply it repairs the auto-fixable cadence/deal-task ones through the single tagged-task owner. It reuses only existing reads, so it's safe to run on demand or on a schedule — and it's what keeps crm worklist trustworthy.

Filtering leads by status uses Close Advanced Filtering (POST /data/search/), not the /lead/ smart-search — Close silently ignores status_id:"..." in that query. crm lead list -s <status>, the reach-out sweep, and the Engaged book all route through it; free-text / name / custom-field lead searches stay on /lead/.

Opportunities (deals)

Once a lead replies, it leaves the cold cadence and is worked as one or more Close opportunities — deals moved through a pipeline's stages (Discovery → Trial → Proposal → Won/Lost). Stage ids are pipeline-scoped and resolved from stage names at runtime, so nothing hardcodes a stat_… id. The pipeline, the Deal Type custom field (Co-Pilot / Fully-Run), and the Engaged lead status are created once by a Close admin.

dogfu crm opportunity pipelines                       # stages + their ids
dogfu crm opportunity create <lead_id> --value 1500 --period monthly \
      --deal-type co-pilot --next "Send proposal" --next-due 2026-07-07
dogfu crm worklist --kind deal                        # deals due today (part of the one queue)
dogfu crm opportunity update <opp_id> --stage Proposal
dogfu crm opportunity next <opp_id> -t "Follow up on contract" -d 2026-07-10
dogfu crm opportunity win <opp_id> --value 1800       # or: lose <opp_id> --reason "..."

Each open opportunity carries exactly one open [dogfu:deal] next-step task — the warm-phase analogue of the cadence task. opportunity next (and --next on create/advance) is the single writer: it closes the prior deal task and opens the new one; win/lose close it. Unlike the cold cadence, the task's content and due date are rep-authored — there's no wait curve. A deal's due next-step shows up in crm worklist --kind deal, and crm reconcile audits the deal invariant (≤ one open deal task per open opp; none on a won/lost or vanished opp) plus dropped balls (an open deal with no next-step task) and stalled deals.

win/lose set only the deal's status. The lead-status move (→ Customer / Not Interested) is left to the caller, because it's a separate, human-meaningful decision.

Publishing a first-audit report

report publish pushes a single self-contained HTML audit to the public reports bucket, served at reports.agentberlin.ai/<domain>. The HTML must carry the book-a-call CTA (https://cal.link/berlin) — it's checked locally before any upload (and re-checked server-side), so a report missing it fails fast and nothing is written. Re-running for the same domain overwrites in place.

dogfu report publish --domain acme.com --html ./acme-audit.html
# -> { "url": "https://reports.agentberlin.ai/acme.com",
#      "domain": "acme.com", "published_at": "2026-06-30T12:00:00Z" }

Programmatic use

from dogfu import Dogfu
dx = Dogfu()                                  # resolves DOGFU_TOKEN / config file
dx.seo.domain_overview(target="acme.com")     # -> DomainOverview
dx.crm.create_lead(name="Acme")

Develop

uv run python tests/smoke.py   # offline checks (normalizers, models, SDK wiring)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

dogfu-0.15.0.tar.gz (117.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

dogfu-0.15.0-py3-none-any.whl (91.2 kB view details)

Uploaded Python 3

File details

Details for the file dogfu-0.15.0.tar.gz.

File metadata

  • Download URL: dogfu-0.15.0.tar.gz
  • Upload date:
  • Size: 117.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.26 {"installer":{"name":"uv","version":"0.11.26","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for dogfu-0.15.0.tar.gz
Algorithm Hash digest
SHA256 8899c31e4f1fa81877502ca6d3d8e7f11e1bbdafe3d921fb78777e26f874d66d
MD5 0c4291aac7bb042c0bb24155f4fd9d06
BLAKE2b-256 6f5fb085944bfb1f6305f508c5df5444b06490edb029fa9025f001a5b4baeac6

See more details on using hashes here.

File details

Details for the file dogfu-0.15.0-py3-none-any.whl.

File metadata

  • Download URL: dogfu-0.15.0-py3-none-any.whl
  • Upload date:
  • Size: 91.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.11.26 {"installer":{"name":"uv","version":"0.11.26","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for dogfu-0.15.0-py3-none-any.whl
Algorithm Hash digest
SHA256 e3c643a81593c4a519f4e1e88bd67c759bf1e67f63e22687dd55f7231cab061c
MD5 72a8dd0266357241752891e742c7512f
BLAKE2b-256 c729e081df59e019ce6645ab9c46dae27bd01a5c5f8eb19229d59b9ab6365244

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page