Agents

Planned

A delivery team, not a single generalist

Most AI workflows start with a generalist agent. TaskForce is building a small team of specialized agents — product, architecture and delivery — each grounded in your Memory, each producing proposals a human approves.

The team

Three roles, the way a strong org splits them

Responsibility is split the way a software organization does it — so no single agent has to be an expert at everything, and no single agent gets the final say.

CPO

the what

Shapes the product: vision, specifications, priorities and acceptance criteria.

CTO

the how

Owns technical coherence: architecture, APIs, data, trade-offs and technical risk.

COO

the when & the risk

Owns delivery: sequencing, capacity, dependencies, milestones, risk and drift.

Three perspectives. One run. No agent gets the final say.

Orchestration

One run, three perspectives

Each role produces an artifact — a spec, an approach, a breakdown — and hands it to the next only after a human approves it. Three perspectives, one accountable chain.

A human approves at every checkpoint — no role hands off on its own.

What's real today

One assistant, grounded in your Memory

The specialized team is the direction. Today, TaskForce ships an assistant grounded in your Brain OS: it retrieves the reasoning behind your system, uses connected tools — and hands consequential actions to a human.

ADR-014 · Postgres over Mongo — chosen
Constraint · EU data residency (GDPR)

Search your Memory

Retrieves the decisions, constraints and conventions behind your system before it answers.

New note→ Memory graph
Saved to graph

Take notes

Writes findings back as knowledge nodes, so what it learns stays in the graph.

Connected tools
GitHubSlackLinear
GitHub· read PR done

Use connected tools

Calls tools exposed over MCP by the systems you've connected — and any external write waits for a human.

The rules

What makes an agent governable

An agent is only as useful as it is accountable. Every agent — the assistant today, the specialized team as it lands — works to the same rules.

Grounded

Uses your architecture, decisions and conventions — not generic guesses.

Specialized

Works within a defined responsibility instead of pretending to know everything.

Governed

Proposes; a human approves. Nothing consequential runs on its own.

Your models

Local through Ollama at zero model cost, or hosted — your call, fast or deep per run.

Every proposal an agent makes waits for a human at a checkpoint — see how sign-off works in Approvals.

A team you direct, not a black box you hope for

Specialized agents, grounded in your context, producing proposals you can review, approve or reject.