Roadmap

Purpose#

This roadmap focuses on one goal: make Fred a state-of-the-art yet simple agentic platform for production teams.

Scope: March 2026 to May 2026.

The strategic differentiator is explicit: governance as a first-class platform capability. Many open-source stacks provide agent loops. Fred focuses on policy-driven control across:

• MCP and tool access
• model routing and profile lifecycle
• prompt lifecycle and rollout safety
• agent catalog governance
• data scope and retrieval boundaries

For terminology used below (Temporal, ClickHouse, KPI, MCP, and more), see the Glossary.

Strategic Priorities#

1. Unified governance plane (core differentiator) Consolidate governance contracts and enforcement across MCP/tools, models, prompts, agents, and data scopes.

2. Benchmarking and observability first Keep improving the benchmark tool and KPI coverage to make performance and quality visible by default.

3. Platform evolution with Temporal and ClickHouse Build on the current production usage of Temporal (already used for ingestion pipelines) and extend it to deep-agent and long-running business workflows. In parallel, adopt ClickHouse as a high-speed analytics store for benchmark and KPI-heavy workloads.

4. ReAct and Graph runtime improvements Continue improving both execution models while converging, when practical, toward shared contracts aligned with the Thales genai_sdk.

5. Configuration management as a core capability Strengthen catalog-based configuration so multimodal agents can be managed cleanly across environments.

6. MCP gateways and validation apps Prototype gateway-based MCP integration (obot-like approach) and enable team admins to deploy validation apps for model drift and embedding quality monitoring.

7. UI revamp and collaboration UX Deliver a major UI revamp to reduce technical debt and, more importantly, revisit the full team experience across libraries, corpus usage, agents, and prompts.

Timeline#

Month 1 - March 2026#

Focus#

• Governance baseline and policy contracts
• Observability and benchmark foundations
• Configuration baseline hardening
• UI revamp foundations

Planned outcomes#

• Governance baseline:
  • Define shared policy vocabulary and scope across mcp, models, prompts, agents, and data.
  • Stabilize model routing contracts (capability, purpose, operation) and baseline phase policies.
  • Define prompt asset lifecycle baseline (ownership, versioning, promotion gates).
• Benchmark tool: expand repeatable scenarios (latency, throughput, tool-call heavy flows).
• KPI layer: improve metric consistency (runtime, model, tool, storage, and queue dimensions).
• Dashboard baseline: define a small “must-watch” operational view for daily tracking.
• Configuration: stabilize catalog patterns (agents, models, mcp) and validate schema strictness for future multimodal support.
• UI revamp: launch the first redesign wave to improve UX for team collaboration across libraries, corpus, agents, and prompts while reducing frontend technical debt.

Success criteria#

• Teams can run the same benchmark suite and compare results across builds.
• Core KPIs are queryable and explain performance regressions.
• Governance domains and ownership are explicit and documented.

Month 2 - April 2026#

Focus#

• Governance enforcement in runtime paths
• Runtime platform upgrade
• ReAct/Graph convergence work

Planned outcomes#

• Governance execution:
  • Wire deterministic policy checks deeper in runtime paths (model selection, tool/MCP gating, data scope constraints).
  • Define first admin-oriented management flows for governed catalogs (agents, models, mcp, prompt assets).
• Temporal: keep ingestion workflows stable in production and extend durable execution to deep-agent and long-running business tasks.
• ClickHouse: introduce or expand analytics storage for KPIs and benchmark traces, with a path to handle large historical growth and data-oriented workloads.
• ReAct + Graph: implement targeted runtime improvements and shared abstractions where it reduces duplication.
• genai_sdk convergence: align contracts best-effort (model/runtime/tool interfaces) without blocking delivery.
• Multimodal config: define first config extensions required for multimodal agents.

Success criteria#

• More execution paths are durable/restart-safe under Temporal.
• KPI and benchmark analytics scale better and are easier to query.
• Policy decisions are observable and explainable in core execution paths.

Month 3 - May 2026#

Focus#

• Governance operationalization and auditability
• Validation and production-readiness loop
• Admin tooling and gateway prototype validation

Planned outcomes#

• Validation apps: allow team administrators to deploy validation workflows on Fred.
  • Model drift monitoring
  • Embedding quality monitoring
• MCP gateways: deliver a concrete prototype and document operational patterns.
• End-to-end quality pass: benchmark + KPI + runtime stability review before next release cycle.
• Prompt management: deliver a cleaner governance-ready model with reusable prompt assets, versioning, and safe validation before rollout.
• Governance operationalization:
  • Expose consistent governance controls in admin UX across MCP/models/prompts/agents/data.
  • Provide audit-oriented visibility for policy outcomes and configuration changes.

Success criteria#

• Admin teams can run periodic validation checks without custom scripts.
• Gateway prototype is usable for real integration testing.
• Release decisions are backed by measurable benchmark and quality signals.
• Teams can govern prompt changes with traceable versions and validation gates.
• Governance controls are usable end-to-end across the five domains.

Notes#

• This roadmap is intentionally execution-oriented and may be reprioritized based on field feedback.
• The objective is continuous, practical improvement rather than one large rewrite.