Agent execution infrastructure

Make every agent execution reproducible.

Ontral captures the observations that shaped an AI-agent execution into a deterministic capsule — so teams can replay failures, find the first divergence, test alternative decisions, and prevent regressions.

Your agent may be nondeterministic. Its execution history should not be.

Request early access Watch replay demo

Two lines to record production

$ pip install ontral

import ontral
ontral.record(app="support-agent")

support-agent · illustrative replayarmed · recording

Live executionrecorded

01model.request

02retrieval.policy_v7

03tool.customer.lookup

04model.response

05tool.refund.execute

06outcome.incorrect_refund

Exact replayserved from capsule

01model.request

02retrieval.policy_v7

03tool.customer.lookup

04model.response

05tool.refund.execute

06outcome.incorrect_refund

cap_demo_8f3a91 · 1,204 observations · root 8f21…e104read-only

provider calls

external tool calls

world mutations

demo capsule · illustrative values

Why traces are not replay

A trace shows the execution. A capsule can reconstruct it.

Traditional observability is valuable for inspecting logs, spans, latency, and errors. Exact replay solves a different problem: reconstructing the historical observations that shaped the execution.

Trace

what you have today

logs
spans
prompt and response
latency
errors

Useful for inspection and operations. It cannot reconstruct the historical universe.

Execution capsule

what Ontral captures

captured observations
tool + network activity
retrieval versions
time + randomness
execution ordering
effect attempts

Useful for reconstruction and replay.

Execution capsule

Capture the observations that shaped the run.

ontral.record() captures supported observations across model, tool, network, retrieval, runtime, state, exception, and execution-ordering boundaries — sealed into one durable, verifiable capsule. Select a class to see what is recorded.

cap_demo_8f3a91root 8f21…e104 · content-addressed

Observation class

Retrieval

Retrieved documents and their exact versions at execution time — preserved after the index or document changes.

content-addressed blobs · replay eligibility requires coverage of the run’s nondeterministic boundaries

Exact replay

Reproduce the failure without repeating the damage.

Replay serves captured observations back to the agent instead of calling live providers and external systems again. It works after the model changed, the documents changed, the database changed — or the provider is gone.

provider calls

external tool calls

world mutations

The original execution was nondeterministic. The captured capsule is deterministic.

ontral · replay — demo capsule

$ ontral replay cap_demo_8f3a91 --attest

capsule restored       1,204 events
provider calls         0
external tool calls    0
world mutations        0
replay status          identical
attestation            verified

Debugger + counterfactuals

Find what changed. Then test the correction safely.

Replay a capsule against current code, prompts, or models. Ontral walks both executions together and stops at the first observation where they disagree — then lets you replace one observation and replay the alternate history, dry-run.

debuggerearly accesscounterfactualsdesign partner

Execution debugger · illustrativecap_demo_8f3a91 ↔ current @ HEAD

nodehistorical · capsulecurrent · HEAD

Selected node · policy.retrieve

The retrieval boundary returned document v8 where the capsule recorded v7. Every later difference is downstream of this observation.

retrieval · first divergence

Counterfactual · replace one observation

policy_v7 → policy_v8

replay from node 09 · code, state, and prior observations stay fixed · live tail stays dry-run

Original · recorded

09 · retrieval.policy — v7

10 · model.response — decision: refund

11 · tool.refund.execute — $1,240.00

outcome.incorrect_refund

Counterfactual · dry-run

09 · retrieval.policy — swap v7 → v8

10 · model.response — decision: escalate

11 · tool.escalate.create — dry-run · no live ticket

outcome.human_escalation

verified · zero side effects

Regression + drift

Turn incidents into tests. Know when behavior changes again.

Promote important production capsules into a regression corpus replayed before every change — and keep replaying a pinned corpus on a schedule to catch drift in models, prompts, code, tools, and retrieval.

PR #482 · retrieval pipeline update

Illustrative CI report · replay gate

412 demo capsules replayed

identical398

semantically equivalent10

changed3

unsafe effect attempted1

Build failedforbidden change class detected

cap_demo_2c11d8 · tool.refund.execute attempted during replay — effect blocked, build rejected

Behavioral drift detecteddesign-partner capability

corpus: refund-escalation · 64 capsules
previous: safe escalation
current: autonomous refund
first divergence: model.response · node 142
trigger: model alias updated upstream

alert → escalation policy reviewoutcome flip

model changes
prompt changes
agent-code changes
tool changes
retrieval changes

Fidelity + evidence

Exactness must be proven — not assumed.

Ontral is designed to fail closed: an uncovered execution boundary must produce REPLAY INCOMPLETE rather than a false identical result. And when a replay does match, the match is cryptographically verifiable.

Replay incompletefail-closed

source        thread-3
operation     socket.connect
coverage      unsupported boundary
remediation   install the relevant interceptor
              or annotate the boundary

designed to fail closed · eligibility reported per capsule

Incomplete capture is a first-class diagnostic, not a silent success.

Illustrative attestation · cap_demo_8f3a91

original root · capture

8f21c39ab7…e104

replay root · attested

8f21c39ab7…e104

Matchtwo roots · one execution · independently verifiable

This is not blockchain. It is verifiable execution evidence.

Integrations

Capture the boundaries that shape execution.

Nondeterminism enters an agent at specific boundaries. Ontral instruments each supported one, records what crossed it, and serves it back on replay.

All execution boundaries →

OpenAIcore technology

Requests, responses, streamed chunks, and tool calls at the client boundary.

Anthropiccore technology

Messages, streaming, and tool use.

HTTPXcore technology

Sync and async requests, responses, and failures.

Requestscore technology

Request and response pairs, including status and errors.

MCPcore technology

Protocol traffic between your agent and MCP servers: calls, results, and attempted mutations.

timecore technology

Clock reads.

randomnesscore technology

Draws and seeds.

UUIDscore technology

Generated identifiers.

async orderingcore technology

Task interleaving as lanes.

tool wrapperscore technology

Internal functions and proprietary tools wrapped as explicit captured boundaries.

Frameworks that execute through supported provider, network, protocol, and runtime boundaries may already be capturable without a dedicated adapter. First-class adapters — LangGraph, CrewAI, OpenAI Agents SDK, internal frameworks — are prioritized with design partners.

Enterprise

Your execution data. Your deployment boundary.

The capture and replay engine is the product foundation. The broader team and enterprise control plane is being productized with design partners — deployment follows your data boundary, not ours.

Ontral Cloud

Hosted capsule store and replay engine.

design partner

Customer VPC

Capsules never leave your network.

roadmap

Self-hosted

Run the full stack on your infrastructure.

roadmap

On-premise

For regulated and air-gapped environments.

roadmap

Identity, RBAC, audit logs, retention, redaction policies, customer-controlled storage, BYOK, and residency are roadmap items built with design partners — none are claimed as generally available today.

Enterprise details →

Category clarity

Agent execution infrastructure. Nothing broader.

Ontral is

execution capture
exact replay
debugging
counterfactual testing
regression infrastructure
drift control
execution evidence

Ontral is not

an agent framework
a prompt-management tool
a generic observability dashboard
an agent marketplace
a model-training platform
an all-in-one governance platform

Observability and replay solve different problems. Keep your traces — add the capsule.

Make every important agent execution reproducible.

Work with Ontral to make production failures replayable, explainable, testable, and preventable.

Apply for early access Become a design partner

Read the product architecture →