A workspace is a scoped, AI-governed runtime — provisioned on your infrastructure or fully managed by us. Turn it into a customer agent, a developer console, a test bed. Policy-checked, audited, traceable from the first call.
One dies when the lid closes. The other is always on, always audited, always duplicable. Same tools. Different foundation.
Every system listed here is owned, deployed, and governed by the platform — not a wrapper around a third-party API. This is how agents improve: we trace, we RAG, we eval, we speed-test. Three tiers — pick the one that matches your ambition.
Every deployment includes all Basic features. Pro and Golden are the full platform — no tier gates, no add-ons.
Not a starter plan — the platform baseline.
Workflows, policy, vault, audit. If an agent runs on ToolShell, it runs governed. Included in every deployment — no gates, no add-ons.
For teams that build with us.
Prompt testing, security scanning, inbound integrations, pod control. The full agent workshop — not a black box.
Agents that earn trust.
Tracing proves what happened. RAG proves what they knew. Evals prove what they got right. Three pillars, one platform.
Workspace → scope → governed. Three steps, same for every customer.
Policy, vault, audit baseline — configured per customer. Every workspace is isolated. Customer A never touches customer B's data.
Agent-only, full console, or both — depends on what your customer needs. Same governance model, different surface. Start small, scale when ready.
No agent runs without a record. Audit trails are immutable. Traces are live. You know exactly what every agent did, when, and for whom.
Three layers. One direction. No hidden state.
Every action taken, deferred, or decided writes a ticket. Auto-linked to workflow runs. Not logs, not Slack threads — the record your team actually reads.
Named, versioned, replayable execution plans. Clone and test a repo. Run a security audit. Onboard a customer. Durably stored, retryable from any failed step.
Scheduled triggers over workflows. Optional — The CLI is always the primary path. When you do need automation, a loop is the only sanctioned trigger in the system.
Six principles that keep your agents governed — no matter what changes underneath.
Anthropic, OpenAI, DeepSeek, a local model — the platform doesn't care. Providers plug into the workflow model; they don't define it. Swap a vendor without touching core logic.
Every platform action has a CLI verb. If ToolShell supports it, the CLI exposes it. No hidden behavior, no GUI-only paths. The CLI is the contract — everything else wraps it.
Services and providers change constantly. We don't rebuild the system when they do. Our job is to make workflow-building flexible, portable, and resilient — not to own the wiring on both sides.
A workspace defines what work can happen, who can do it, and what it can touch. It is not an agent. It is not unlimited access. It is a controlled, governed scope.
Tickets are ground truth. Workflows define what needs to happen. Loops define what needs to repeat. Work flows in one direction — no hidden state, no ephemeral decisions.
Simple, complex, manual, automated, multi-agent — the same cascade model handles all of them. No special cases, no rigid patterns forced on teams. The model bends; the principles don't.
Services are workspace-scoped data feeds and sinks. They don't carry identity inside the platform. They are pluggable inputs and outputs, not participants in governance.
| Service | Role | Direction |
|---|---|---|
| GitHub | Code source, PR triggers, repo sync | source + sink |
| Jira | Ticket read/write from workflow results | source + sink |
| Confluence | Documentation source for RAG ingestion | source |
| Slack | Workflow notifications and operator alerts | sink |
| Datadog | External observability and metric feeds | source |
| Telegram | Operator bot, gate reports, chatbot surface | sink |
See the live demo — provision an agent, watch the execution plan go green, copy the client config. Every call audited.
Book a demo →Our in-house solution for non-deterministic R&D — a governed workshop where human engineers and AI agents share the same CLI, the same probes, the same runbooks. Bootstrap a tenant, onboard integrations, run API/UI/CLI test flows, and review evidence bundles — all from one deterministic command layer with an agent-readable skill surface.
Explore the Blueprint →