deepline-gtm-agent

Open-source GTM chat agent powered by Deepline. The default path is Deepline v2 native agent/chat: your app brokers Slack, REST, and web chat requests while Deepline handles tool routing, enrichment, research, CRM actions, and provider-specific workflows through the v2 API.

API portal: code.deepline.com - create your DEEPLINE_API_KEY there.

Quickstart

cd managed_agent
pip install -r requirements.txt

export DEEPLINE_API_KEY=dlp_...

python server.py     # starts REST, web chat, and Slack endpoints on :8000

Open http://localhost:8000 for web chat, or call the REST API:

curl -X POST http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "find emails for 5 VP Sales at fintech companies"}'

What it does

The agent handles common GTM workflows with Deepline's v2 tool catalog and API:

Workflow	Example prompt
Contact enrichment	"Find the email for Jane Smith at Acme"
Prospect search	"Find 10 VP Sales at B2B SaaS companies, 200-500 employees, US"
Account research	"Research stripe.com and summarize GTM-relevant signals"
Email verification	"Is jsmith@acme.com safe to send?"
LinkedIn resolution	"Find the LinkedIn URL for Tom Nguyen at Notion"
CRM and outreach	"Create a HubSpot contact" or "show my Lemlist campaigns"

Responses should include sources, provider outcomes, and a clear next step. The agent should state data gaps instead of inventing missing emails, titles, or company facts.

Eve reference implementation

This repo now includes an additive Eve reference implementation in eve_agent/. It preserves the Deepline v2 execution backend while using Eve for durable sessions, local HTTP, evals, and fast Vercel deployment.

Use it when you want an out-of-the-box deployable agent path:

cd eve_agent
npm install
npm run link
# set DEEPLINE_API_KEY in Vercel env or local .env.local
npm run dev
npm run smoke -- --host http://127.0.0.1:3000

See eve_agent/README.md for the full local, eval, and Vercel flow.

Architecture

Slack / REST / Web UI
      |
      v
FastAPI broker
      |
      v
Deepline v2 agent/chat + SDK/API
      |
      v
Deepline integrations, enrichment providers, CRM, outreach, and research tools

Configure access with environment variables and call the Deepline v2 SDK/API directly. Managed sessions should not depend on local Deepline CLI state.

Boundary

This repo should stay thin. It owns:

• REST, web chat, and Slack transport
• Slack request verification and formatting
• optional bearer auth for public chat endpoints
• CORS and deployment setup checks
• prompt/tool bounds that steer requests into Deepline

Deepline API owns provider routing, plays, workflows, enrichment, CRM/outreach actions, credentials, billing, run state, and workflow observability. Do not copy those systems into this agent.

Hermes Compatibility

This repo also includes hermes-agent-pack/, the compatibility layer for running the Deepline GTM agent inside Hermes on a persistent Sprite/Fly-style workspace.

Use it when Hermes is the operator interface and Deepline is the GTM execution, logging, workflow, and observability layer. The pack makes the Hermes setup explicit:

• pruned Deepline context, claims, exclusions, and Jai voice rules
• Hermes prompts and skills for one primary deepline-gtm-agent
• bounded subagent workflows for sales, account research, CRM hygiene, AgentMail, proof review, and workflow specs
• split marketing specialists for content, campaign planning, and proof/claims review
• Telegram, AgentMail, connector, and spawn-k2qb setup docs
• the HTML deck for the Hermes AI marketing team call recording

Start with hermes-agent-pack/README.md, then run hermes-agent-pack/prompts/00_seed_hermes.md in Hermes.

Run the shared eval suite against a Hermes profile with:

python tests/run_evals.py \
  --hermes-command "deeplinegtm -z" \
  --output tmp/hermes-eval-results.json

For the Sprite-hosted profile:

python tests/run_evals.py \
  --hermes-command "sprite exec -s spawn-k2qb -- deeplinegtm -z" \
  --output tmp/hermes-sprite-eval-results.json

Interfaces

Web chat

Run python managed_agent/server.py and open http://localhost:8000.

Setup checks

Use /doctor to verify non-secret deployment configuration:

curl http://localhost:8000/doctor

The response reports missing auth, wildcard CORS, Slack setup, and unsafe local/live-write combinations without returning API keys or tokens.

REST

curl -X POST http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Research rippling.com"}'

With endpoint protection enabled:

curl -X POST http://localhost:8000/chat \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{"message": "Find 3 VP Sales in the US"}'

Slack

Set SLACK_BOT_TOKEN and SLACK_SIGNING_SECRET, then DM the bot or mention it in a channel. See SETUP.md.

SDK/API

Use DEEPLINE_API_KEY for Deepline v2 API calls. Keep API keys in environment variables or your deployment secret store.

import os
import httpx

resp = httpx.post(
    "https://code.deepline.com/api/v2/integrations/apollo_search_people/execute",
    headers={"Authorization": f"Bearer {os.environ['DEEPLINE_API_KEY']}"},
    json={"payload": {"job_title": "VP Sales", "limit": 5}},
    timeout=60,
)
resp.raise_for_status()
print(resp.json())

For full chat behavior, use the v2 agent/chat SDK or API from the broker layer instead of shelling out to local CLI state.

Deploy

See SETUP.md for Railway and Slack setup. Required production variables:

Variable	Required	Description
DEEPLINE_API_KEY	Yes	Deepline v2 API key
PORT	Yes	Usually 8000
API_KEY	Optional	Protects /chat endpoints with bearer auth
CORS_ORIGINS	Optional	Comma-separated allowed origins; empty disables browser CORS
SLACK_BOT_TOKEN	For Slack	Slack bot token
SLACK_SIGNING_SECRET	For Slack	Slack request signing secret
REDIS_URL	Optional	Persistent Slack thread history

ANTHROPIC_API_KEY, MANAGED_AGENT_ID, and MANAGED_ENVIRONMENT_ID are only needed for the optional Anthropic Managed Agent shell in managed_agent/setup.py; they are not required for the default native Deepline v2 broker.

For the Eve on Vercel path, see eve_agent/README.md and the Vercel section in SETUP.md.

Legacy self-hosted agent

The root Python package contains a legacy self-hosted agent path for local experimentation. It is not the recommended deployment path. New deployments should use the v2 native agent/chat flow above.

License

MIT