MeerTech
First autonomous deployment in production.
MeerTech is an execution layer — the agent runs against your live systems of record, under your policy envelope, with every action recorded. This walk-through gets your first agent from workspace provisioning to shadow mode to live autonomy.
Three layers, one execution system.
Every MeerTech deployment composes the same three layers. Understanding what each layer owns is the fastest path to designing a working agent.
Execution — plan, tool-call, retry, escalate, commit
Intelligence — unified live context agents reason over
Trust — policy gating, identity, lineage, audit
Your systems of record, telemetry, documents, people
Vocabulary.
Agent
A scoped, policy-bounded autonomous process with a role, a set of allowed tools, and a trace of every action it takes.
Context
The live operational state agents reason over — systems of record, documents, policy, history — unified by Neural Grid.
Tool
A callable capability an agent can invoke. Registered under Sentinel with a policy envelope — who, what, when, how.
Policy envelope
The bounded rule set that gates every autonomous action: scope, limits, approvals, data residency, escalation path.
Trace
The immutable audit of every reasoning step, tool call, policy decision, and outcome for a given agent run.
Escalation
Hand-off to a human operator when a threshold, risk, or policy rule triggers. First-class in MeerTech, not a fallback.
Six steps to production.
Every MeerTech deployment follows the same sequence. Shadow mode is not optional — it is how agents earn the right to act.
01
Provision a workspace
Request access via the deployment flow. MeerTech stands up a workspace in the environment you've chosen — managed cloud, private VPC, sovereign region, or hybrid.
02
Connect your systems
Register Neural Grid connectors for the systems of record the agent will read from and act on. Each connector is scoped by role and routed through Sentinel.
03
Define the policy envelope
Declare what the agent may do autonomously and where the human gates are: dollar limits, touched entities, required approvers, data residency, escalation SLAs.
04
Register tools
Tools are the agent's hands. Each is declared via the SDK with input/output schema and a policy binding. MCP-compatible services can be registered directly.
05
Ship the agent plan
Write the plan in Python, TypeScript, Go, or Java. Deploy with one command. Agentic Core handles scheduling, retries, state, and long-horizon orchestration.
06
Run in shadow, then live
First runs execute in shadow mode: Sentinel records every decision the agent would have taken without committing. Graduate to live autonomy once policy and trace review is clean.
Hello, autonomy.
The minimum working agent. A policy-bounded reconciler that ingests invoices, checks them against a budget envelope, and posts the cleared journals via the Zoho Books MCP connector.
agents/reconciler.py
python
from meertech import Agent, policy, tool
@tool(provider="mcp.zoho_books", action="post_journal")
def post_journal(entry: dict) -> str: ...
agent = Agent(
name="finance.reconciler",
role="finance.agent",
policy=policy.envelope(
scope=["post.journal", "reconcile.bank"],
limit_usd=50_000,
approvers=["controller.queue"],
escalate_above_usd=25_000,
data_residency="NG",
),
tools=[post_journal],
)
@agent.plan
def close_period(period: str):
for invoice in agent.context.invoices(period=period):
if agent.sentinel.allows("post.journal", invoice):
post_journal(invoice.to_entry())
else:
agent.escalate(invoice, reason="above autonomous envelope")shell
bash
$ meer agent deploy agents/reconciler.py --workspace primary --mode shadow → workspace=primary env=shadow role=finance.agent ✓ policy envelope validated ✓ tools registered (1) ✓ neural grid connectors resolved (4) → shadow run started. view at meer.app/runs/rn_9f2a
Who does what.
MeerTech deployments have three distinct operating roles. Each has its own surface, its own permissions, and its own audit trail.
Operator
Runs agents in production. Owns the runtime dashboards, approval queues, and escalation response.
Review escalation queue
Approve / override autonomous actions
Pause or scope agents live
Inspect decision traces
Developer
Builds and deploys agents. Writes plans, registers tools, defines policies, ships new capabilities.
Author agent plans and tools
Register MCP connectors
Configure policy envelopes
Ship via the SDK / CLI
Compliance / Auditor
Owns the governance surface. Reviews policies, pulls audit reports, satisfies regulator queries.
Query audit lineage
Review policy changes
Export compliance reports
Sign off deployment gates