We ship production AI you can actually depend on.

bold.black is a consulting studio for production AI. We design and ship agentic software, automate the workflows that drain your team, and embed senior engineers when you need to move now.

Start a project What we do

/ AI software development / Agentic engineering / Workflow automation / Staff augmentation

Services

Four ways we move your roadmap forward

We work across the full lifecycle — strategy, build, and the boring-but-critical production engineering that makes AI actually reliable.

AI software development

End-to-end product engineering with LLMs at the core — from RAG pipelines and evals to the UI your users actually touch.

RAG, retrieval & vector search
Model evaluation & guardrails
Full-stack product delivery

Agentic engineering

Autonomous and human-in-the-loop agents that plan, call tools, and complete real work — built to be observable, testable, and safe.

Multi-step tool-using agents
Orchestration & state management
Tracing, evals & observability

Workflow automation

We find the repetitive, high-cost workflows hiding in your operations and replace them with reliable, monitored automation.

Process discovery & mapping
Integrations across your stack
Monitoring & failure handling

Staff augmentation

Senior AI and platform engineers who plug into your team, ship in your codebase, and level up your people while they're there.

Embedded senior engineers
Architecture & code review
Knowledge transfer built in

Products

Fleet

internal platform Visit site

A self-contained agentic development environment

Fleet is a sovereign platform where AI agents do real engineering work — write code, run CI, and ship to Kubernetes — entirely on infrastructure you own, with zero public endpoints.

We built Fleet to answer the question every serious team is now asking: how do you put autonomous AI agents to work without handing your code, secrets, and infrastructure to someone else's service? Fleet packages an entire software organization — source control, CI/CD, chat, workflow orchestration, and the agents themselves — into a single reproducible EKS cluster that bootstraps from one CloudFormation template and a set of agent skills. Everything runs behind a private network. Because security and data isolation are architectural defaults — not bolt-ons — the same design maps cleanly onto HIPAA, SOC 2, and similar regulatory frameworks.

Sovereign by design

Code, models, and data stay on infrastructure you own and can audit. After hand-off, the platform even hosts its own source of truth — no external dependency on day two.

Agents that actually ship

Hermes agents don't just chat — they read the repo, push commits, trigger CI, and deploy through GitOps. The full engineering loop, running autonomously.

Reproducible & disposable

Stand the whole environment up — or tear it down — from declarative infrastructure and gated agent skills. No snowflake servers, no manual runbooks.

Compliant by construction

Private networking, full audit trails, and infrastructure you own and control are built in from day one — so Fleet is designed to satisfy HIPAA, SOC 2, and similar audits by its nature, not retrofitted to pass them.

// stack EKS · GitOps · agents

EKS Auto Mode
self-provisioning Kubernetes, no node ops
ArgoCD GitOps
declarative, self-healing workloads
Tailscale
private, encrypted access — no public endpoints
Forgejo + Actions
self-hosted Git & CI with container builds
Hermes on Bedrock
AI agents that operate over Matrix chat
Temporal
durable workflow orchestration

Products

Corkboard

coming soon

A hosted knowledge base for AI-augmented teams

Corkboard is a hosted wiki built for how AI-augmented teams actually work — a shared, searchable brain that both humans and agents read from and write to throughout the day.

Corkboard captures what scattered chat threads, agent transcripts, and stale docs cannot: a durable, interlinked knowledge base that every agent reads from and writes back to — readable pages for people, a clean API for agents. Where Fleet gives your agents infrastructure and Dispatch puts them in your channels, Corkboard gives them memory.

Memory that survives sessions

Agent context disappears when a session ends. Corkboard is the durable store — what an agent learned about your codebase, your conventions, your decisions — written to a page the next session reads on its first turn.

Same page, same data

There is no export step between what a person reads and what an agent queries. A page a human edited this morning is the same object an agent cites this afternoon — one source of truth for both.

Auditable by construction

Every edit carries attribution and lands in a revision history you can walk back. You can trace what an agent wrote, when, and why — the record is the audit log.

// stack managed · API · MCP

Built for dual authorship
humans write pages, agents write them too — same data, one place
Full-text + semantic search
find pages by keyword or by meaning
Agent-native interfaces
a clean HTTP API and an MCP server out of the box
Revision history
every change tracked, attributed, and reversible
A graph, not a folder
pages cross-link both ways — context travels with the page
Fully managed
sign up, get a workspace, never touch a server

Open source

Harness

MIT licensed Visit site View on GitHub Documentation

Easy containerized agent environments

Harness runs coding agents inside a hardened, sandboxed container — point one at a project and let it work without handing over your whole machine.

We open-sourced the sandbox we wanted for ourselves. Harness wraps Docker around three open-source coding agents — pi, opencode, and hermes — behind a single CLI. Every run is capability-dropped, signature-verified, and supply-chain hardened, so you get the upside of autonomous agents without the blast radius. It runs locally against LM Studio by default, or any major cloud model with one flag.

Agents, contained

Each run is locked to a single mounted directory in a capability-dropped container with no-new-privileges and a seccomp profile — the agent never touches the rest of your machine.

Trust the supply chain

Images are signed and verified with cosign + SLSA provenance on every run, and dependencies sit out a 7-day cooldown to dodge freshly-published supply-chain attacks.

Your models, your call

Local-first with LM Studio out of the box, or drop in a key for Anthropic, OpenAI, OpenRouter, Gemini and more — same CLI, same flow either way.

// features MIT · npx · docker

Sandboxed by default
cap-drop ALL, no-new-privileges, seccomp
Three agents, one CLI
switch pi · opencode · hermes with -a
Supply-chain hardened
cosign + SLSA provenance, 7-day dep cooldown
Local-first
LM Studio by default, any cloud model with -e
Stateful or one-shot
persist sessions or run fully ephemeral
Zero install
npx @boldblackai/harness just works

try it

$ npx @boldblackai/harness -p "write a fizzbuzz in Go"

Open source

Dispatch

MIT licensed Visit site View on GitHub Documentation

Long-running coding agents that live where your team is

dispatch spins up your own long-running dispatch agent — an opinionated hermes-agent deployment — inside your Slack and Teams channels, running entirely on an AWS account you own.

dispatch is a CLI that scaffolds a complete repository for one long-running agent, each with its own Slack and Teams identity. Generate a @swe-pal for code reviews and pull requests, or a @reporter that posts scheduled reports to the channels you choose — create, customize, and deploy as many as you like. Every dispatch agent runs hermes-agent on our hardened harness image inside one isolated ECS cluster, with Slack, Teams, and GitHub integration built in, state persisted and backed up on EFS, and everything deployed into infrastructure you control.

A dispatch agent in your workspace

Each dispatch agent is one long-running hermes-agent with its own Slack and Teams identity — spin up a @swe-pal for code reviews and PRs, or a @reporter for scheduled updates, right where your team already works.

Production-shaped from day one

Every dispatch agent ships on the same hardened foundation we run ourselves: hermes-agent on one isolated ECS cluster, Slack + Teams + GitHub wired in, and state persisted and backed up on EFS — no glue code, no half-finished infrastructure.

Your account, your control

Everything deploys into your own AWS account — a CloudFormation stack, scoped IAM roles, KMS-encrypted SSM, and one isolated ECS cluster — generated, managed, and torn down with gated skills.

// features MIT · npx · aws

Lives where your team is
an opinionated hermes-agent deployment in Slack and Teams
Runs on your own AWS
one isolated ECS cluster in an account you control
Slack, Teams + GitHub built in
reviews code and opens PRs where your team works
Persisted & backed up
state on AWS EFS survives restarts and updates
Your account, your control
CloudFormation, scoped IAM, KMS-encrypted SSM
Skill-driven lifecycle
scaffold, /setup, /manage, /teardown — one dispatch agent per repo

try it

$ npx @boldblackai/create-dispatch swe-pal

How we work

A short path from idea to production

No bloated discovery phases. We de-risk fast, ship the real thing, and leave you with something your team can run.

01
Scope
A focused working session to pin down the problem, the constraints, and what 'done' looks like. You leave with a concrete plan, not a sales deck.
02
Prototype
We build the riskiest slice first — a working prototype against your real data — so we prove value before anyone commits to the full build.
03
Ship
Production engineering with evals, observability, and CI baked in. We deploy into your stack and hand over something your team can own.
04
Scale
Monitoring, iteration, and knowledge transfer. We make ourselves replaceable — or stay on as the embedded team you can scale with.

Start a project

Tell us what you're building

Drop the details below. We read every message and reply within one business day — usually with a few sharp questions and a sense of how we'd approach it.

Senior engineers only — you talk to the people building.

Fixed-scope projects or embedded engagements.

We'll tell you if we're not the right fit.