Below are the exact scenarios we built and ran — the problem, the signal we sent, and what the system produced. Copy the commands. Boot it up in your own environment. Then tell us honestly what you found.
CRM has the records. It doesn't have the relationship. When an AE walks into a renewal call, they're piecing context together from six tabs — email threads, Slack history, call notes, deal room — while the customer expects them to already know everything. The signals exist. They're just not connected.
# Download the ZIP, then: $ cd contextecf-fabric-trial/ $ cp .env.example .env $ docker compose up # Migrations run first. Ready when you see "Listening on :8080". # First pull: 2–3 min. After that: ~30 seconds. # Verify the platform is healthy: $ curl http://localhost:8080/health/ready | jq # Ingest the renewal meeting signal (mock-nce.json is included): $ curl -X POST http://localhost:8080/v1/context/events \ -H "Authorization: Bearer trial-dev-token" \ -H "Content-Type: application/json" \ -d @mock-nce.json | jq
renewal_risk, champion_engaged, and upsell_opportunity map to signal types your team actually tracks?
sourceSystem, sourceEventId, capturedAt. Every signal is attributed. Can you replace google_calendar with your actual source system and still get a clean receipt?
Every team ships their own LLM integration. Anthropic for some, OpenAI for others, Google for a third. Costs are invisible until the invoice. There's no policy layer. When something goes wrong, no one knows which model did it, under what instructions, with what data. You can't govern what you can't see.
code_generation),
selected the optimal model tier using RunwayClass routing, estimated cost at the declared call volume,
and returned the full decision rationale — before making a single inference call.
# Copy the config template (add a key when ready): $ cp gateway.yaml.example gateway.yaml $ docker compose up # Ready in ~20 seconds. Redis included. # See governance decisions — no API key needed: $ curl -X POST http://localhost:8090/v1/preflight \ -H "Content-Type: application/json" \ -d '{"estimated_calls":500, "task_domain":"code_generation", "context_tokens_avg":4000}' | jq # Once your key is set — send a governed inference request: $ curl -X POST http://localhost:8090/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{"model":"claude-sonnet-4", "messages":[{"role":"user", "content":"Summarise our Q1 pipeline risk"}]}' | jq
estimated_calls and watch it scale.
AI agents are making decisions — drafting emails, querying databases, calling APIs. When a compliance officer asks "who authorized this action, under what context, with what constraints?" the answer today is "we think the model did it." That's not an audit trail. That's a liability.
previous_receipt_hash references the first —
forming a tamper-evident chain, persisted to a local volume,
with no external dependencies, no cloud account, no trust-us.
# Start ACM in standalone mode (no Pub/Sub, no Fabric needed): $ cd acm-trial/ $ docker compose up # Ready in ~20 seconds. Redis included. # Cycle 1 — governed agent activation: $ curl -X POST http://localhost:8095/v1/agent/cycle \ -H "Content-Type: application/json" \ -d '{"trajectory_id":"demo-traj-001", "task":"Summarise Acme renewal history"}' | jq # Cycle 2 — same trajectory, watch the chain grow: $ curl -X POST http://localhost:8095/v1/agent/cycle \ -H "Content-Type: application/json" \ -d '{"trajectory_id":"demo-traj-001", "task":"Draft renewal email for Alice at Acme"}' | jq # Inspect the receipt chain in the Control Tower: $ curl http://localhost:8095/v1/admin/control-tower | jq
docker volume inspect acm_outbox after the cycles. Receipts persist even if the container restarts. Does that match your air-gap or data residency requirements?
The SDK and appliance are designed to be embedded — in your agent runtimes, at your network edge. Docker scenarios coming in the next design partner cohort.
The TypeScript SDK that coding agents and AI runtimes embed to receive governed activation. Owns the contract types — ActivationReceipt, RuntimeEnforcementManifest, DeniedCapabilityReceipt — plus context compiler and negative knowledge.
Standalone appliance that runs at the edge — destination guard decisions, runtime manifest enforcement, masking, denied-capability receipts, and SIEM export. Evaluates every governed action against the active manifest before it executes.
Design partners get 180 days free on any tier. No credit card. No auto-renewal. Tell us what you think at the end.
Collected from design partners who ran these scenarios in their own environments. No staging accounts. No curated demos.