I
AgentInk
Launch Agent
Self-Evolving Slack Developer Agent

The autonomous developer that lives in your Slack

AgentInk is a self-healing, proactive parent agent that orchestrates specialized child agents, writes and tests code, facilitates live meetings, and integrates deeply with your entire engineering stack.

agentink-live-development.sh
deploy_e2e: @agentink build a new child agent for customer support and link it to #support-channel
<system> On it...
✦ AgentInk:

I have successfully created and registered the child agent "support-bot".

Configuration:

  • ID: support-bot
  • Channels: #support-channel
  • Status: active
deploy_e2e: @agentink run Claude Code to fix the database connection leak in the backend repo (ticket LIN-402)
✦ AgentInk:

Executing Claude Code session in secure Docker container...

✓ Cloned repository aiinkwell/inkwell-backend

✓ Created branch feature/LIN-402-db-leak

✓ Applied fix & ran unit tests (all passed)

✓ Opened Pull Request #142: "Fix connection pool leak in PostgreSQL client"

Engineered for Modern Teams

AgentInk isn't just another chatbot. It's a fully integrated, self-healing developer platform operating directly inside your workspace.

Self-Healing Loop

Detects runtime errors, writes code changes, triggers local test suites, and commits verified fixes autonomously.

Parent-Child Agents

Spawn, configure, and monitor specialized child agents for support, QA, or deployment directly from Slack.

Android Telemetry

Run autonomous tasks on Android emulators using ADB and Gemini vision, complete with live dashboard streaming.

Deep integrations with the tools you already use

AgentInk connects directly to your engineering ecosystem, allowing you to manage tasks, documents, physical security, and meetings without leaving Slack.

L

Linear Integration

List, create, complete, and sync tasks directly to your Neo4j Knowledge Graph.

M

Meeting Facilitation

Invite AgentInk to Google Meet, Zoom, or Teams. It introduces itself, walks the agenda, and posts minutes to Slack.

U

UniFi & IoT Control

Monitor network devices, view camera snapshots, and unlock doors securely with human-in-the-loop approval.

Active Integrations

LN
Linear
GD
Google Drive
NT
Notion
UF
UniFi Protect
N4
Neo4j Graph
MS
Moonshot API

Stateful & Secure Architecture

Built on a robust, enterprise-grade foundation designed for maximum reliability and strict data isolation.

REASONING ENGINE

Gemini 3.5 Flash

Powered by Google's state-of-the-art gemini-3.5-flash via the native google-genai SDK.

KNOWLEDGE GRAPH

Neo4j Database

Maintains a rich, queryable graph of tasks, meetings, people, and system dependencies.

DATA SECURITY

PostgreSQL RLS

Row-Level Security ensures encrypted per-user secrets and conversation history remain strictly private.

DEPLOYMENT

Single GCP VM

Runs as a single systemd service (inkwell.service) on GCP, communicating via Slack Socket Mode.

Verify before you speak

Built to police itself

A continuously-curated benchmark, always-on telemetry, and an anti-fabrication pipeline keep AgentInk honest — and prove it with numbers.

Gold-Standard Eval Curation

A benchmark that can’t drift. Questions are generated deterministically from the live Neo4j graph and supplemented with human-curated edge cases — every snapshot versioned, quality-gated, and immutable.

  • Deterministic + human-curated + edge cases
  • Checksum-versioned snapshots, published only on change
  • Quality gates: min rows · required types · no dupes
  • Rebuilt hourly from Linear → Neo4j

System Health & Telemetry

Always-on observability. A live dashboard tracks knowledge-graph growth and structure, evaluation quality, agent latency, and token cost — with scheduled checks that alert Slack the moment something drifts.

  • Graph metrics: nodes/edges · density · orphans · cycles
  • Latency p50 / p95 / p99 + token cost
  • Workspace Completeness Index
  • 6-hour checks · nightly snapshots · Slack alerts

Verify-Before-You-Speak

It checks itself before it answers. A veracity gate matches every claim against the turn’s tool ledger, an LLM judge scores faithfulness on sampled and flagged turns, and a nightly adversarial suite hunts for fabrication.

  • Veracity gate on every turn (claims vs. evidence)
  • LLM-as-judge: 10% of turns + 100% of flagged
  • Nightly false-premise probes → fabrication_rate
  • 3-pass offline eval: retrieval · faithfulness · correctness

Strict Development Workflow

AgentInk enforces a rigorous Order of Operations to guarantee code quality and system stability.

1
Proposal

Discuss in Slack

2
Branch

Create feature branch

3
Test

Local unit & smoke tests

4
PR & Review

Open GitHub PR

Live & self-hosted

Deploy your autonomous developer to Slack

AgentInk runs as a single service in your own cloud, talks to your team in Slack, and ships real work — code, websites, meetings, and more.