Hector coordinates a team of specialised AI agents that plan, build, validate, and integrate software — at machine speed, with humans supervising, not operating.
No spam. Early access when it's ready.
You're on the list. We'll be in touch.
Existing tools make individual developers faster. They don't solve what happens when you want agents to work together — reliably, at scale, without a human in every loop.
Every output requires a human to review, approve, and pass on. You're not automating a team — you're automating an inbox.
Deciding what to include in each agent's prompt is manual, inconsistent, and expensive. Too little and agents make mistakes. Too much and you waste tokens on irrelevant detail.
Agents have no memory of what worked, what failed, or what the codebase's conventions are. Each invocation re-learns what the last one already knew.
Hector encodes the full propose–validate–integrate cycle as a machine-speed protocol. Agents negotiate with each other. Humans step in for genuine judgement calls — not routine reviews.
Proposals rejected by QA go back to the engineer with structured feedback. Up to three revision rounds before a task escalates to a human supervisor.
Hector ships with the infrastructure that makes autonomous collaboration reliable — so you don't have to build it yourself.
A propose–validate–integrate protocol replaces code review as a bottleneck. Agents submit structured proposals; a QA agent returns typed verdicts. Accepted work is committed automatically. Rejected work comes back with specific, actionable feedback.
Each agent invocation receives a tailored context payload — relevant source files ranked by import relationships, type matches, and file mentions, trimmed to the token budget. Agents get what they need, nothing they don't.
Every task execution is logged — outcome, iterations, rejection reasons, what worked. Pattern-level insights are distilled over time. Agents learn the codebase's idioms, its fragile modules, and which approaches succeed for which kinds of work.
Architectural decisions are first-class entities — linked to the project graph, surfaced automatically in agent context, and append-only. Agents respect settled constraints without anyone having to remember to mention them.
Every agent output is typed, versioned, and tracked with provenance and consumer metadata. When an upstream artefact changes, the platform knows exactly which downstream work is affected — Git can't tell you that.
Configurable approval gates for deployments, architectural changes, and scope decisions. Role-based capability matrix, SHA-256 integrity checks, and a full structured audit log. Autonomy with visibility, not instead of it.
You watch the system operate and intervene when it needs steering. Hector's supervisory interface gives you real-time visibility into every agent, every task, and every decision — without requiring your attention at every step.
Hector is open source under the Apache 2.0 licence. Inspect it, run it, extend it. Hosted services coming later for teams that want the infrastructure managed for them.