Groclake · the enterprise agent harness
Groclake is the agent harness for real enterprise work: Nutane, the assistant your teams talk to; claws that do the work in sandboxes; one command line to every system; and the models, tools and data underneath. Agent RBAC, encrypted agent-to-agent calls and a full audit trail are built in.
Built by Plotch.ai
user → nutane → claws → gash → groclake-cli → groclake → modellake · toollake · datalake
The Groclake Harness
From the assistant you talk to and the agents you write down, to the models, tools and data they run on. Every tier is usable on its own and every request is governed end to end.
T0 · Assistant
Agentic Harness Assistant
Users talk to Nutane by chat or voice. It plans the work, runs background tasks, creates and manages claws, asks for approval and uses every tier below.
T1 · Agents
Command-line agentic workers. Groclake users and customers create, deploy and manage their own agents, each in its own sandbox.
T2 · Scripts
Groclake bash. Scripts that compose groclake-cli commands with gash-specific commands into repeatable workflows.
T3 · Interface
The Groclake command line: one scriptable interface to every resource and service on the platform.
T4 · Framework
The framework underneath. Provisions resources for Groclake agents, runs hardwired agents and manages every database type.
T5 · Services
Model management and routing, tool management with 50+ enterprise connectors, and context, data and document management.
T0 · Nutane
Your teams ask in plain language, by chat or voice. Nutane plans the work, finds the right tools, runs long jobs in the background and asks before anything risky. Every screen below is the real product.
Nutane guard
Read only, Can make changes, or Auto with the guard. Reads go straight through; anything that changes data, touches a shared system or works in bulk stops for an approval card with the reason on it.
Claws from chat
"Every Monday at 9 am, email me…" becomes a scheduled, sandboxed claw. You see exactly what will be created and approve it first.
Memory
Say "remember that…" and Nutane keeps it across chats. Every memory is visible and editable, and sensitive personal data is never stored.
Skills
Teams package their know-how as skills. Type / to see them; Nutane loads only what the task needs.
Context
Instructions, memory, conversation, tool results and loaded tools, each counted, so context is never a black box.
Real work, real data
It finds the right commands, runs them against your systems and answers with tables, sources and caveats, not guesses.
And more, all governed by the same guard
T1–T3 · Claws, Gash, groclake-cli
Claws are command-line agentic workers. Gash scripts compose groclake-cli commands into repeatable workflows. Everything Nutane can do, your engineers can script.
Every claw runs in its own sandbox with its own service identity, never with a user's raw credentials.
Cron schedules, one-off background runs, or synchronous calls when an answer is needed now.
Any claw can be published as a tool that Nutane and other claws call, with the same permission checks.
One claw spawns sub-claws in parallel, with heartbeats and stale-run recovery, and merges their results.
$ groclake claw create weekly_triage "Triage open incidents"$ groclake claw schedulesweekly_triage Mon 09:00 IST next in 2d 4h$ groclake memory list --scope user$ groclake skill install --skill-id market-digest$ groclake qops algo bell run$ groclake gash run gash/ops/nightly_triage.gash
T4–T5 · Groclake framework and the lakes
The Groclake framework provisions resources for agents, runs hardwired Groclake agents and manages every database type. The lakes are where agents reach models, enterprise systems and data, with the same permissions and audit everywhere.
Model management and routing. Pick per task, with budgets and usage logs per account.
50+ enterprise app connectors, ready as tools for Nutane and claws. Credentials stay in tool connections, never in prompts.
Context, data and document management. Agents read structured, unstructured, vector and graph data through one access layer, with row-level account checks.
Structured
Relational stores such as MySQL, PostgreSQL and Snowflake.
Unstructured
Documents, files and records in MongoDB, S3, SharePoint and Drive.
Vector
Embeddings in pgvector, Elastic and Mongo vector search.
Graph
Relationships in Neo4j for people, assets and dependencies.
Security and governance
Agents act on real systems, so every call is authenticated, authorised, encrypted and recorded. Your security team can see exactly what each agent did and why.
Tight role-based access for every agent, user and intent, with tenant isolation between accounts. Agents get only the permissions their role grants.
Every agent-to-agent call carries a verified identity. Tokens rotate, and claws sign in as their own service accounts.
Calls between agents travel encrypted, and sensitive resources are encrypted at rest with rotatable keys.
Admin rules allow, ask or deny by command. Risky, bulk and shared-system actions wait for a person.
Every agent call, guard decision, approval and claw run is logged with who, what and why, for agents and the harness alike.
Agent traces, model usage and cost, failures and activity reports, per account and per agent.
Enterprise use cases
Nutane spawns as many sub-claws as a job needs, runs them in the background and merges every result, with approvals where they matter. A few of the workflows teams are building:
IT services
Signals from Datadog and Dynatrace, incidents in ServiceNow, one sub-claw per client running known runbooks. P1s wait for approval and go to on-call in Teams.
Ports & logistics
Tomorrow's arrivals from SAP compared with berth plans, manifests and customs papers, one sub-claw per vessel. Conflicts reach the port desk before the ship arrives.
HR & workforce
CVs parsed, matched against open roles on skills, location and fit, and anyone with an incomplete profile nudged by email or SMS on a schedule.
E-commerce
Shopify orders watched for sales velocity, a purchase order drafted in SAP for approval, and the listing updated when stock lands.
Quantum · QOps early access
Circuits, pulses and algorithms through the QOps library, run on IBM Quantum, Amazon Braket or Azure Quantum from the same harness.
Launch film
88 seconds: the guard, memory, skills, claws from chat, 50+ connectors, sub-claw orchestration and the use cases above.
Deploy anywhere
One Helm chart, the same harness on-prem or in your cloud, with any mix of models.
Run the whole harness in your data centre, with self-hosted open-source models and no data leaving your network.
Kubernetes-native with a single Helm chart, on AWS, Azure or Google Cloud, scaling out per agent.
Route each task to foundation, open-source or self-hosted models, with per-account budgets and usage logs.
Built by Plotch.ai
Plotch.ai bridges the translational gap between business problems and deeptech: we diagnose, formulate and execute, using qualified talent and Groclake, our proprietary engine. Groclake is the harness we run our own work on, now available for your enterprise.
Groclake
The enterprise agent harness
Nutane · Claws · Gash · groclake-cli · Lakes
We will show you Groclake running it, with your systems, your models and your security rules.