Product Roadmap

Building the Trust and Policy Layer for Autonomous Agents

Timeline: What we're building from Q4 2025 to Q4 2026

The Product Suite

🛠️ CLI: Protocol validation and local guard development

🛡️ Guard Service: Go-based runtime engine for policy enforcement

🌐 Platform: Agent registry, policy control plane, and observability

Shipping Today: RFC001–003(Agent Control Protocol, Trust Badges, Key Ownership Proofs)

RFC004 Draft

RFC005 Draft

RFC006–008 Draft

🎯

Design Partner Program

Want to influence what we build? We are working with a small group of teams already running agents in production or staging environments.Get direct input, priority access, and dedicated support.

Apply as a design partner →

Q4 2025 – Foundation & Core Engine

Focus: Stabilizing the Go enforcement core and developer tooling.

CLI v1.0 Stable Release

Full agent envelope validation and signing
Local evaluation of trust policies (uses the same engine as production)
JWS/JWKS signature verification and debugging

Guard Service (RPC) – Alpha / Internal

•Initial Go-based guard service exposing the core engine over local RPC
•Unix Domain Socket / localhost TCP transport for low latency
•SDK prototypes for Node.js and Python calling the guard via RPC
Note: This is an internal milestone, not a public GA feature.

Q1 2026 – Platform Launch

Focus: The single source of truth for agent identity and security posture.

CapiscIO Platform Launch

•Agent & Server Registry: Identity and integrity store for A2A agents and MCP servers
•Event Log: Structured event storage for badge verification, guard decisions, and agent activity
•Visual Dashboard: Trust posture metrics, agent inventory, event timeline, and API key management

Integrations & SDKs

•Unified SDKs: Node, Python, and Go libraries powered by the local RPC Guard
•LangChain & MCP Integrations: Native trust enforcement for LangChain/LangGraph chains and MCP tool servers
•Starter Tier: Free tier for individual developers and small agent swarms

Q2 2026 – Cross-Org Zero Trust

Up Next

Focus: Agent A from Org A calls Agent B in Org B — Org B sets the rules. Cross-organization policy enforcement built on badge identity.

Cross-Org Trust Policies

In Development

•Inbound Agent Policies: Org B defines rules for agents calling in from Org A — allow, deny, or scope based on badge claims, org identity, and capabilities
•Observe Mode: Evaluate policies without blocking — log decisions for tuning before enforcing
•Zero-Trust Defaults: Every cross-org call requires a valid badge; no implicit trust between organizations
MVP: Basic cross-org policy with allow/deny rules per badge claim — demo-ready for design partners.

Enterprise Management

•Team Workspaces: Role-based access control for policy management
•SIEM Export: Push decisions and violations directly to Splunk, Datadog, or Sentinel
•SSO/SAML: Enterprise identity integration

Q2–Q3 2026 – TCHB & Observability

Focus: Transaction binding and understanding the "Why" behind agent decisions.

Transaction & Hop Binding

RFC004

•TCHB Implementation: Bind multi-hop agent workflows into verifiable chains of custody
•Hop Tracing: Trace caller → agent → sub-agent chains via signed hop attestations
•Trust Graph Visualization: Explore agent relationships and delegation paths in the dashboard (built on TCHB data)
Note: Trust graph and delegation chain visualization require RFC-004 TCHB as a prerequisite.

Observability Integrations

•Event Streaming: Export guard decisions and policy evaluations to your existing stack
•Native exporters planned for Datadog, Prometheus, and Honeycomb

Q3 2026 – Delegated Authority & Policy Engine

Focus: Full delegation stack and policy enforcement — building on TCHB (RFC-004) shipped in Q2–Q3.

Delegated Authority & Full Policy Engine

RFC005

RFC008

Authority Envelopes (RFC-008)

•Delegation Chains: Cryptographic hash-chained envelopes — Agent A delegates scoped authority to Agent B with monotonic narrowing
•Capability Classes: Dot-notation scoped capabilities (e.g. finance.payments.read) with prefix-based subtree inheritance
•Enforcement Modes: Observe → Guard → Delegate → Strict — progressive tightening per envelope issuer

Policy Engine (RFC-005)

•PEP / PDP Contract: Engine-agnostic policy evaluation — compatible with OPA, Cedar, or custom engines
•Constraint Narrowing: Anti-escalation enforcement — child delegation can never exceed parent scope
•Obligations & Break-Glass: Mode-specific obligation handling and signed emergency override tokens

Agent Graph Explorer

•Delegation Chain Viewer: Visual tool to explore agent relationships and delegation paths built on TCHB + envelope data
•Policy Decision Log: Searchable history of PDP allow/deny decisions with full context

Enterprise Deployment

•On-Premise Guard: Run the Guard Service in air-gapped or private cloud environments
•SOC 2 Type II: Compliance certification (in progress)

Q4 2026 – Analytics, Marketplace & Automated Response

Focus: Reputation intelligence and the "Visa" network for agents.

Reputation & Analytics

•Reputation Scoring: Risk and trust scores for agents, publishers, and relationships
•Outlier Detection: Detect anomalous chains of authority based on graph heuristics
•Compliance Reports: Exportable audit trails for regulatory requirements

Verified Agent Marketplace

•Exploring a directory of agents with verified identity and reputation scores (demand-dependent)
•"Verified by CapiscIO" badges based on cryptographic proof history

Automated Response

•Kill Switches: Policy-driven containment for high-risk or compromised agents
•SOAR Hooks: Trigger automated incident response workflows in external security tools

Available Today

Published packages you can install and use right now.

capiscio (npm)

The official CLI for validating A2A agents. Wraps the Go engine with a Node.js interface.

npm install -g capiscio

Commands:

validate – Agent Card validation
key gen – Ed25519 keypair generation
badge issue/verify/keep – Trust Badge management
gateway start – HTTP reverse proxy with badge validation

capiscio (PyPI)

Python-packaged CLI wrapping the same Go core engine. Install via pip, run from any Python environment.

pip install capiscio

Same commands as the npm version. The Go binary is downloaded automatically on first run.

capiscio-sdk (PyPI)

Runtime security middleware for Python agents. Includes SimpleGuard for JWS signing/verification.

pip install capiscio-sdk

Drop-in middleware for FastAPI, Flask, or any Python agent framework.

langchain-capiscio (PyPI)

Alpha

Trust enforcement for LangChain and LangGraph agents. Drop a guard into any LCEL chain.

pip install langchain-capiscio

Composable via the | pipe operator: CapiscioGuard() | ChatOpenAI()

capiscio-mcp (PyPI)

Beta

MCP server identity and tool-call attestation. Secures Model Context Protocol tool servers.

pip install capiscio-mcp

Provides MCPServerIdentity for signing and verifying MCP tool calls.

capiscio-core (Go)

The core validation and scoring engine. Embeddable in any Go application.

go get github.com/capiscio/capiscio-core

Key packages: badge, simpleguard, did, pop, gateway, scoring, mcp, revocation, and more.

validate-a2a (GitHub Action)

CI/CD integration for agent validation with three-dimensional scoring.

uses: capiscio/validate-a2a@v1

Outputs: compliance-score, trust-score, availability-score, production-ready.

Quick Install

Node CLI:npm i -g capiscio

Python CLI:pip install capiscio

Python SDK:pip install capiscio-sdk

LangChain:pip install langchain-capiscio

MCP:pip install capiscio-mcp

Go:go get .../capiscio-core

Vote on Features

How to influence the roadmap

Open a GitHub discussion

Share your use case and desired capability.

Community reacts

Other developers and partners upvote and add context.

We prioritize

We review top-requested items with design partners and fold them into the roadmap where they fit.

Submit Feature Request

Design Partner Benefits

Higher Priority

Design partners' requests are evaluated first and often used to seed new prototypes.

First Access

Priority access to internal builds and prototypes.

Special Terms

Priority support and favorable commercial terms when hosted services become available.

Apply as a Design Partner

Build the future with us

CapiscIO today is a CLI, Python SDK, and Go engine you can ship with.
The roadmap is about turning that into a full trust and policy layer for autonomous agents.

Start with CLI Apply as a Design Partner