Excapia by Tracston

Define the mission. Excapia finds the way.

Turn business objectives into adaptive MissionFlows. Execute across applications and infrastructure, recover within your rules, and verify the outcome with evidence.

Continuous business journey assurance. Adaptive execution. Verified outcomes.

Explore MissionFlow Request a Demo

Interactive product demonstration

A MissionFlow for “Verify checkout after deployment” Plan leads to Execute. The recorded step fails, so execution takes an alternate route, then Verify checks three conditions and the objective resolves as verified with evidence. Objective Checkout verified ✓Plan6 steps ✕Executeselector changed ↷Recovervalidated route ✓Verify3 assertions

MissionExample

Mission
Verify checkout after deployment
Scope
Staging
Mode
Supervised
Success
  • Order exists
  • Payment is confirmed
  • Confirmation is delivered
  1. 01 The mission
  2. 02 The path
  3. 03 The boundary
  4. 04 The proof
01 · The mission

Start from the outcome, not the script.

A MissionFlow begins with what the business needs to be true, written as success conditions that can be checked.

Every step can finish and the business outcome can still be wrong. A checkout page that loads, a button that clicks, a job that exits zero: none of them prove that an order exists, that it was paid, or that the customer heard about it. Excapia keeps those two things apart: step status and mission outcome.

Objective
Verify checkout after deployment
  • Order exists in the order storeassertion
  • Payment is confirmedassertion
  • Confirmation is deliveredassertion

Step finished

“Clicked Place order.” Says nothing about the order.

Outcome verified

Order, payment and delivery each confirmed with evidence.

Success conditions are part of the mission, not an afterthought.
02 · The path

Repeatable when it can be. Adaptive when it must be.

Most of the time, a deterministic workflow is exactly right: fast, cheap, reviewable, and identical on every run.

Excapia runs recorded and designed steps deterministically, with checkpoints, retries and pause/resume, without calling a model. When a step fails in a way retries cannot fix, an adaptive branch can investigate and propose a different route, but only inside the boundaries you set. A validated, recurring recovery can then be turned back into a deterministic step.

  • Deterministic execution without model calls
  • Checkpoints, retries, pause and resume
  • Agent investigation and repair proposals
  • AI-to-deterministic conversion of verified paths
OpenCart Orderfails Investigateadaptive branch VerifyObjective
DeterministicAdaptive branchPaused original path
The objective does not move. The route does.
03 · The boundary

Autonomy operates inside your rules.

Every mission carries its permissions, budgets, environments and approval points, and they are visible before anything runs.

A privileged action outside policy is blocked, not attempted. An action that needs a person pauses the mission in Waiting for approval with the evidence and risk in front of the approver. Declining is a valid outcome: the mission stops truthfully as Blocked instead of pretending.

  • Approval as an engine action and Approval Center
  • Policies, mission contracts and dry runs
Blocked

Restart production database
Not in this mission’s allowlist. Not attempted.

Approval required

Roll back inventory-service to 2.13.4?

Why
Stock lookups fail on 7 of 10 requests since 2.14.0
Action
Allowlisted rollback · privileged
Scope
Staging cluster
Permissionsread-only in production
Budget3 retries · 1 repair
Environmentstaging, test hosts only
Verification3 assertions required
Illustration of an approval boundary. Try it live in scenario B below.
04 · The proof

A completed run should come with proof.

When the mission resolves, the graph becomes an evidence pack: what was asserted, what was observed, when, and with which artifacts.

Screenshots are redacted and stored with the run. Assertions name the system they checked. A repaired step never marks a mission verified on its own; only passing assertions do.

  • Redacted screenshot evidence
  • S3-compatible evidence store
  • Video and trace evidence
Evidence pack · run 7f3cVerified
Order exists
order store · ORD-20481		Passed
Payment confirmed
payment API · PAY-58213		Passed
Confirmation delivered
test mailbox · 10:41:19		Passed
Recovery applied
selector repair · approved		Recorded
checkout.pngconfirmation.pngorder.jsondecision-log.json
Illustrative evidence pack.

MissionFlow Theater

The objective stays. The path adapts.

Give Excapia an outcome, define the boundaries, and follow execution from the first action to verified completion.

Interactive product demonstration · Illustrative demo data · Runs locally in your browser and contacts no system

Scenario A: checkout after deployment. This walk-through runs in your browser with JavaScript. Without it, here is what happens:

  1. The objective is defined with three success conditions: the order exists, payment is confirmed, the confirmation is delivered.
  2. A browser worker follows the recorded checkout path in a fictional staging store.
  3. A storefront release renamed the checkout button, so the step fails. Deterministic retries are exhausted.
  4. An agent examines the screenshot and proposes a replacement selector, which is validated inside the configured scope and approved.
  5. Execution continues on the alternate route.
  6. Order, payment and delivery evidence are checked. The mission is Verified only because all three assertions pass.

Journey Assurance

Know your business journeys work, not just your servers.

Excapia continuously runs the journeys your customers depend on and tells you when the business outcome breaks.

Quote to payment to issued policy. Search to basket to order. Each journey has checkpoints that mean something to the business, baselines that show what normal looks like, and release gates that stop a regression before it ships.

Journey Assurance
  • Browser journeys

    Real Chromium with redacted screenshots.

  • API checks

    HTTP steps with business checkpoints.

  • Baselines

    Rolling P50 to P99 and success rate per journey.

  • Release gates

    PASS, WARN, BLOCK from journey evidence.

  • Desktop paths

    Windows and Linux application journeys.

Recorder to Mission

Show it once. Review it. Assure it forever.

The semantic recorder captures what you meant, not where you clicked, and turns it into a mission you can review.

  1. Capture

    Record a sign-in and quote. Excapia stores intent, target and expected result; test inputs become variables. Credential fields are blocked.

  2. Review

    The recording becomes a draft. You see each business action and can rename, reorder or remove it before anything runs.

  3. Workflow

    Approved drafts become versioned, repeatable journeys with schedules and datasets.

  4. Success assertions

    Add the conditions that prove the outcome, such as “policy issued” or “order exists”. Now it is a mission.

Native agent

An agent that works for the mission.

The Excapia agent receives the mission, the relevant platform context, its tools, the evidence so far and its boundaries. It is a participant in the MissionFlow, not a chatbot beside it.

AI Assist drafts missions and edits designs as reviewable changes, using models you host. When a step fails, the agent investigates, proposes a repair and works with coding and multimodal tools, always inside the mission’s boundaries.

Meet the agent
  • AI Assist

    Mission drafts and canvas edits, kept as drafts until reviewed.

  • Experience and Skills

    Learn from verified recoveries; reuse proven skills.

  • Native execution loop

    Durable progress, tools, scheduling and bounded recovery.

  • Coding and investigation

    Repository, diff, terminal and tests inside boundaries.

  • Multimodal evidence

    Screenshots, audio and sampled video frames where supported.

Inside Excapia

Built for operators who need to know why.

Real Excapia screens, shown with synthetic demo data.

Mission Control for “Deliver the daily settlement file to the partner”: state Completed, budget, attempts, one adaptation, three completion assertions passed, and a decision panel showing the switch from SFTP to the partner API fallback.
Mission Control

Objective, state, budget, adaptations, completion assertions and the decision behind the current strategy. Synthetic demo data.

MissionFlow Designer: a primary SFTP strategy with two fallbacks, independent verification and a verified outcome node, with the mission contract’s outcome, success criteria and boundaries on the right.
MissionFlow Designer

Strategies and fallbacks, independent verification, success criteria, allowed and forbidden actions. Synthetic demo data.

Journey executions table with business journeys, healthy or failed outcomes, versions, durations and seeds.
Execution history

The exact path, version, outcome and seed of every business validation.

Recorder capturing a sign-in on a synthetic motor insurance portal, with recording options and capture settings.
Recorder

A controlled remote browser records business actions on a synthetic insurance portal.

Continuity and Recovery screen with counters for stalled, recovering, blocked, waiting and recovered missions.
Continuity & Recovery

Stalled, recovering, blocked and waiting missions, with the engine’s recovery decisions.

Agent Studio: mission conversation, boundary, task plan, a selector diff, sandbox terminal and passing tests with a Request approval button.
Agent Studio

Conversation, task plan, diff, terminal, tests and artifacts, inside the mission’s boundary.

Continuity

Keep working when the link drops. Never on stale authority.

Work survives restarts and outages because progress is checkpointed and journaled, and reconnection reconciles before anything new starts.

Checkpoints, stall detection and a bounded recovery pipeline keep missions moving. A branch worker that loses its link continues only the checks its authority lease allows, waits for fresh approval before any change, and stops when the lease expires.

  • Step checkpoints and resume
  • Stall detection
  • Bounded recovery pipeline
  • Disconnected execution with leases

Branch worker Confirmed

Checkout journey passed · terminal check passed (journaled)

Central view Stale since 10:42:13

Knows the checkout journey passed. Nothing after the outage.

Price file update Waiting

Needs fresh central approval. Cannot proceed offline.

Illustration of scenario C. Run it in the MissionFlow Theater.

Deployment and ecosystem

Runs on your infrastructure. Works on its own.

Excapia runs as a private Docker stack on infrastructure you control. Models are endpoints you configure. Tracston integrations are optional.

Private Docker stack

Server, workers, Chromium, database, queue and evidence storage in one Compose project.

ExampleA staging VM running the full stack with all features on, used for demos and QA.

Customer-managed models

AI features call endpoints you run, such as Ollama. Public demo input on this site is never sent to a model.

ProvidersOllama, vLLM-compatible servers, Claude, Gemini and OpenAI, each an explicit configuration choice.

Tracston integrations

Optional: Observatory for telemetry and investigation context, Imperium for AI governance, Otopia for agent collaboration.

IndependentExcapia operates fully without any of them.

Deployment options

Use cases

Missions teams start with.

Verify checkout after every storefront release

Assert
Order exists, payment confirmed, email delivered
Adapts
UI changes in recorded paths, within scope
Gate
Release blocked on failure

Prove a deployment is serving customers

Assert
Health, startup, a real sample journey
Adapts
Allowlisted rollback after approval
Gate
Fresh verification after recovery

Keep an insurance quote-to-policy journey healthy

Assert
Quote priced, payment taken, policy issued
Adapts
Business alerts tied to the failed run
Gate
Baseline regression warning

Deliver a file to a partner, whatever the channel

Assert
Partner confirms receipt; checksum matches
Adapts
Falls back from SFTP to the partner API
Gate
Claims rejected unless verified

Your mission

Bring one outcome you need to be sure of.

We will model it as a MissionFlow with you, run it, and show you the evidence.

Request a Demo Explore all scenarios