From the creators of Dapr and KEDA

Keep agentsrunning.Prove everyaction.

Automatic recovery for agents, MCP servers, and workflows - with cryptographic proof of execution, lineage, and identity.

Start for free Book a demo

Free to start · enterprise & self-hosted available

Same agent. Same crash. Different outcome.

built with

LangGraph

Without Diagridrunning

Retrieve context

Call LLM

Execute tool

Update memory

Final answer

Attempt #1 · no progress kept $10

With Diagridrunning

Retrieve context

Call LLM

Execute tool

Update memory

Final answer

Resumed · 0 re-runs$10

Diagrid is trusted by leading companies worldwide

01Durable execution

Never lose progress.
Pick up where you left off.

Your apps never have to start over. Diagrid preserves execution state as work progresses, enabling automatic recovery from failures without losing progress.

Automatic crash & outage recovery
Exactly-once steps — no duplicate paid calls
Long-running work, from seconds to infinity
Durable state across restarts and scale events

run #8a2f1cstate persisted

Plan0x91a4

Retrieve context0x3c7d

Call toolcrashed

↻ resumed — Call tool0xf02e

Update memory0xb6a4

Respond

0 work lost · 0 duplicate LLM calls

02Verifiable execution

Don't trust logs.
Prove execution — cryptographically.

Every step, decision, and tool call is signed into a tamper-evident record. Know exactly who executed what, where it came from, and prove it wasn't altered — for audits, compliance, and trust.

Provenance — where every action originated
Attestation — who executed each step
Tamper-evident, cryptographic audit log
End-to-end distributed tracing

provenance chain verified

Origin

user · req 7f2

init

Agent step

signed

0x91a4

Tool call

signed

0x3c7d

Altered record

hash mismatch

0x????

Tamper detected — record rejected

Production-ready in two steps

Make every app durable and governed.

01Durability

Make failures and outages irrelevant

Resume from the last completed step
Survive crashes, deploys, and infrastructure outages
No rewrites, no new orchestration platform to learn

agent.pyPython

# Define your LangGraph agent as usual
graph = StateGraph(State)
graph.add_node("process", process_node)
graph.add_node("validate", validate_node)
graph.add_edge(START, "process")
graph.add_edge("validate", END)
graph = build_graph()
compiled = graph.compile()
# Make the agent durable with automatic recovery
runner = DaprWorkflowGraphRunner(
    graph=compiled,
)
runner.start()

then

02Zero-trust access

Control which agents, apps, and MCP servers can talk

Deny by default — allow each connection explicitly
Govern apps, tools, MCP servers, and agents
Enforced at the execution layer, not bolted on

mcp-policy.yamlYAML

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: mcp-policy
spec:
  accessControl:
    # Phase 1 — strict zero trust. No agents, no allows. Reject everything.
    defaultAction: deny
    trustDomain: public