Run AI actions in Cells. The rest stays in your cloud.
Secure, isolated execution infrastructure for AI agents and workflows. Short-lived cells, with policy frozen at birth.
[ Made for your stack ]
Create Workflows
One Flow, Every Run
Turn agent flows into something ops can run and audit.
Explore workflows
$ blekline workflow list
wf_order-triage act · blekline-medium
wf_support-draft infer · local
wf_media-export act · blekline-large
$ blekline workflow use wf_order-triage
$ blekline workflow describe wf_order-triage
nodes decide (your stack) → act (blekline cell)
default cell profile: blekline-medium · ttl 15m
compute blekline-runtime (team) · region eu-central-1
$ blekline workflow validate --dry-run
✓ policy bound · act eligible · region eu-central-1
workflow ready — blekline workflow run wf_order-triageRun cell instances
Cells, Not Sandboxes
When a workflow needs to act, Blekline spins a short-lived cell.
Explore cell runs
$ blekline workflow run wf_order-triage
routing decide (your stack) → act (blekline cell)
runId run_8f2k9m
compileHash a3f9c2e1…82fc frozen at birth
$ blekline run describe run_8f2k9m
tier act · node tool
profile blekline-medium · 2 vCPU · 4 GiB · ttl 15m
compute blekline-runtime (team) · eu-central-1
status pending → assigning cell
$ blekline run watch run_8f2k9m
T_boot 1.2s · job/cell-8f2k9m-7xq assigned
status running · act-min 0:04 · tools 3
run active — act side effects under compileHash a3f9…82fcEnforce actions
A Decision on Every Action
Allow, mask, or block at the MCP hop, before the model moves on.
Learn about enforcement
$ blekline proxy status
hop agent → mcp-proxy → policy → downstream
surface cursor · workspace 234jkk904l
$ blekline evaluate run_shell
tool run_shell
args cmd: rm -rf /var/agent-data
scan secrets 0 · lineage ok · tier high
→ BLOCK policy:destructive-shell-deny
audit evt_9c2a · metadata-only · forwarded: false
$ blekline mask prompt
decision mask_and_send · entities 2
masked … [EMAIL_002] … [AWSACCESSKEY_001][ Where Blekline fits ]
Between your Agents and your Stack.
[ Use cases ]
Pull requests, refunds, and the tools you already run. Those actions belong in a cell.
Let agents open PRs, run checks, and promote releases from act cells. Every tool call stays in policy with a run record ops can review.
Run support and billing agents in short-lived cells. Mask sensitive fields in chat, and block replies or refunds that policy does not allow.
Give SRE agents guarded restarts and rollbacks inside act workflows. High-risk steps can hold for approval, with evidence on each action.
Connect CRM and warehouse tools through governed cells. Query and sync on policy, with masked prompts and metadata-only audit for data teams.
Automate grants and offboarding with policy frozen at run start. Agents cannot widen access without rules you set, and each step lands in the audit log.
Split schema work into infer vs act steps. Heavy migrations and backfills run in isolated cells so blast radius stays bounded and compliance can trace runs.
Questions Platform and Infra Teams Ask
FAQs
Cells & Workflows
01
What is a Blekline Cell?
A short-lived execution instance for workflow act steps: tools, side effects, and runtime policy bound when the cell starts. It is not a long-lived VM or a generic chat sandbox. Workflows describe the path; cells run the parts that need isolation and a frozen policy snapshot.
02
How is this different from “just use a sandbox API”?
Sandboxes isolate compute. Blekline adds workflow + policy: decide on your stack, act in a cell, with compileHash / policy frozen at birth, run IDs, and audit metadata ops can tie to a workflow. You can still use your sandboxes behind act nodes; Blekline is the run and audit layer on top.
03
What are infer vs act in a workflow?
Infer stays where you already run models (local, your API, your GPU). Act is when the workflow needs governed execution (tools, APIs, refunds, schema jobs), that path routes into a Blekline cell with a profile (blekline-medium, region, TTL). One flow; only the act leg spins infrastructure.
cloud & Kubernetes
04
Where does Blekline run?
Act runs on Blekline-managed runtime in your chosen region (e.g. eu-central-1), wired to EKS, AKS, or GKE in your estate. Inference and most of your stack stay in your cloud. Blekline does not ask you to move the whole agent platform.
05
Do we have to replatform our agents?
No. You define workflows and wire clients (Cursor, CI, internal agents) to the same CLI/SDK/MCP patterns you use in dev. Existing decide/infer paths stay; you add validate → run → watch for act when a workflow needs a cell.
06
What do platform teams operate day to day?
Profiles (size, TTL), regions, workspace boundaries, and run lifecycle: list/describe/watch runs, compileHash on each run, metadata for audits. Enforcement at the MCP hop (allow / mask / block) is optional on the same plane, not a separate product silo.
Policy, enforcement & proof
07
What does “policy frozen at birth” mean?
When a cell run starts, Blekline captures a compileHash (and bound policy) for that run. Side effects execute under that snapshot, so ops and security can answer what policy was in force for run X.
08
Does enforcement slow engineers down?
Designed as infrastructure defaults: validate before run, decisions at the tool-call boundary (allow / mask / block), then the cell runs. Teams adopt one workflow first; open-core components let security review the path without blocking every IDE.
09
What gets logged?
Metadata-first audit by default: run and workflow IDs, tool names, allow/mask/block decisions, timing. Not a full prompt dump unless you configure otherwise. Export-friendly for platform reviews; your RoPA, retention, and DPIA stay your responsibility.