Imladri is the runtime control layer between autonomous agents and real-world side effects: tools, payments, code, databases, and compute. It is now proven across OpenClaw, Hermes, MCP, and custom agent runtimes, governed database branches, and third-party GPU runs.
published constitution v7
payment action selected
denied before side effect
hash-chained packet ready
policy -> request -> decision -> halt -> exportPick the framework, hosted workflow, MCP host, or CI lane your team already uses. The full matrix, proof packet, authority path, and scanner details are one click away.
Certified adapter paths for the frameworks teams already use.
Certification lanesCurrent SDK matrix passed with zero failed lanes.
Hosted adopter proofsDify, Flowise, n8n, Zapier MCP, and Botpress proof paths.
Product surfacesAuthority tokens and GitLab/Vercel scanner proof are separate surfaces.
The same rule repeats across agents, databases, and third-party GPUs: decide before the side effect, record what happened, and leave behind an artifact an operator can inspect.
OpenClaw, Hermes, MCP, and custom HTTP runtimes blocked dangerous tool bodies before customer code ran.
A 5B-row Postgres source was branched for governed agent work with zero source mutations.
The same H100 allocation ran protected and raw Qwen 7B LoRA; the protected path attested, zeroized, and stayed inside run variance.
Best first pilots are irreversible or high-trust actions: deploy, refund, transfer, database mutation, production config change, or customer-data export. The first outcome is a proof packet a buyer can inspect.