| The job | Collect web content reliably and keep what was collected: the files, the records, and the evidence of how they were produced. | Answer and reach customers on the right channel, with permission, and know afterward exactly what was sent, what came back, who wrote it, and why each send was allowed. | Keep systems in their intended state without giving automation open-ended power: notice what is wrong, apply only a remedy that is allowed, and call it fixed only when a fresh observation shows that it is. |
|---|
| What it produces | Markdown and schema-shaped JSON, typed artifacts and media, exports, extraction manifests, and immutable dataset versions you can query, diff, and cite. | Support conversations with recorded provenance, governed communications, delivery history, and an append-only audit trail. | Durable incident and decision records, verification and stabilization results, recovery and escalation history, an audit trail of every grant and action, and redacted evidence packets. |
|---|
| Your infrastructure | Your proxy pools (BYOP) and, if you use AI, your own AI provider, with credentials encrypted for their owner. | Your Twilio, SendGrid, and Telegram accounts (BYOK), and optionally your own AI provider or model, with credentials encrypted per workspace. | The systems you connect, and only the corrective actions you allow on them. The engine runs registered actions with fixed arguments, never an open shell. |
|---|
| Policy before execution | Identity, ownership, target, egress, abuse, admission, and resource limits before any job runs; page text stays untrusted for every model. | Workspace role, consent and suppression, Trust and Safety, rate and usage limits before any provider call; AI never bypasses them. | Standing policy, a single-use grant per action, ownership fencing, and a budget on how many changes one incident may make—before anything changes. |
|---|
| Evidence | Artifacts, extraction manifests, job history, dataset provenance and version diffs, and citations to the records an answer used. | Support conversations with reply provenance, normalized delivery history, and an append-only, exportable audit log. | Incident and decision history, verification and stabilization results, recovery records, an audit trail, and redacted evidence packets. |
|---|
| AI | Your own AI provider extracts to your schema and answers with citations; a planner proposes jobs that run only when you confirm. | Planned: optional and bring-your-own, working through the same consent, governed actions, and audit as people. Communications works without it. | An escalation inside the controller: AI may name one action from the set you allow, which then passes the same authority, budget, and verification as any remedy. |
|---|
| Operators | Telegram operator surfaceBuilt | Web console to set up and operate; Telegram for daily workBuilt | Operator acts exist in the engine; how customers reach them is not decided yet |
|---|
| Software | ApexAPIBuilt | ApexAPIBuilt | ApexAPIPlanned |
|---|
| AI agents | ApexMCPBuilt | ApexMCPPlanned | ApexMCPPlanned |
|---|
| Operator plugin | Scrawl PluginIn development | Communications / Twilio PluginPlanned | —Not applicable DevOps is an operations product, not a plugin for one. |
|---|
| Metering | Hosted work reserves credits before it runs and settles them after, on an immutable ledger; your AI provider bills you for its own usage. | A monthly subscription; usage measured on an internal ledger with hard limits, never a credit balance to buy. | Planned as paid access to a managed service. Plans and metering are not set. |
|---|
| Hosted service | Not publicly hosted | Not publicly hosted | Not publicly hosted |
|---|
| Pricing | Hosted work metered in credits; your own proxy costs stay with your provider.Pricing coming soon | A monthly ApexAi subscription; Twilio, SendGrid, and any AI provider you bring bill your own accounts.Pricing coming soon | Planned as paid access to a managed service; nothing is priced yet.Not yet priced |
|---|