Colony

Colony is a command interface for B2B go-to-market teams. A founder or GTM lead types a plain-English instruction at /overview and a coordinated set of specialist agents handles the rest: finding and scoring prospects, drafting personalised outreach across five messaging angles, extracting signals from recorded calls, generating post-call summaries, producing content tied to pipeline attribution, and assembling an 8-asset onboarding kit the moment a deal closes.

The platform runs on a shared memory layer called the Knowledge Core — a 10-domain pgvector store that every agent reads before generating output and writes to after completing a task. This means each agent turn is grounded in accumulated institutional knowledge rather than starting from scratch. Pipedrive stays in sync bi-directionally throughout, so the CRM reflects what the agents know and vice versa.

Colony is built for small GTM teams — typically two to ten people — who are running high-complexity outbound without the headcount to manage each workflow separately. Every outbound message and every prospect batch requires human approval before anything is sent. Circuit breakers enforce per-org rate limits. The system is designed to extend a small team's reach, not to operate without oversight.

GTM teams at early-stage B2B companies typically run their revenue motion across five or more disconnected tools: a CRM, an outbound sequencer, a meeting recorder, a content calendar, and a spreadsheet for onboarding. Each tool produces output the others cannot consume. Call recordings sit unprocessed in Google Drive. ICP scoring happens by intuition. Onboarding assets are assembled manually after every closed deal. The coordination tax — context-switching, copy-pasting, chasing stale CRM records — consumes hours that should go toward selling.

The underlying capability to automate each of these workflows already exists. LLMs can qualify prospects, personalise outreach, extract meeting signals, and draft onboarding documents. The gap is orchestration: there is no single surface that connects these capabilities into a coherent pipeline, shares memory across them, and keeps a human in the loop at the right moments.

• GTM Orchestrator: A streaming chat interface at /overview that accepts plain-English instructions and routes tasks to specialist agents, returning results in the same session without requiring the user to switch tools.
• Prospect and Qualification Agents: Pull candidates from Pipedrive LeadBooster, Google Places, SerpAPI, and Unipile; score each against a configurable ICP; apply a hard anti-ICP pre-filter before any sequence spend is committed.
• Message-Gen Agent: Produces outbound message variants across five angles — signal-led, pain-led, referral-led, pattern-break, and insight-led — for T1 through T5 sequence steps, with all drafts held in an approval queue before send.
• Recording Intelligence Agent: Ingests Gemini Meet Notes from Google Drive, extracts structured signals, patches the relevant Pipedrive fields, and commits a vector embedding to the Knowledge Core for use in future agent turns.
• Onboarding Agent: Triggers automatically on Closed-Won and generates an 8-asset Deployment Kit — pitch deck, one-pager, case-study shortlist, pricing reference, rollout calendar, stakeholder map, risk register, and KPI dashboard — written to cloud storage under signed URLs.
• Analytics Agent: Delivers a daily brief each morning covering pipeline state, outbound queue status, HOT replies, and alerts, surfaced both at /overview and via email.

Enterprise buyers are consolidating their tool stacks. A sales team that previously paid separately for a CRM, a sequencer, a call intelligence platform, a content tool, and an onboarding system is now asking whether one orchestration layer can replace all five. LLM inference costs have dropped to the point where running eight specialist agents per GTM motion is economically viable at the deal sizes early-stage B2B companies target — a threshold that did not exist until recently.

At the same time, compliance and security requirements are tightening. Procurement teams at mid-market and enterprise accounts are disqualifying platforms that store third-party credentials in shared infrastructure. Colony's dual secret-store architecture — platform-level keys in GCP Secret Manager and per-org credentials in a KMS-encrypted vault — directly addresses this. Multi-channel outreach is also a growing requirement: LinkedIn reply rates for technical buyers have overtaken cold email, and tools that remain email-only are losing ground. Colony's phase-two roadmap includes a multi-channel sequence engine, sender mailbox rotation, and email deliverability warm-up, all already in active development with test scaffolding and batch runner artifacts present in the repository.