RUNTIME ENGINE
The Runtime Engine for AI Agents
Sinaptic® DROID+ deploys production AI agents in days, not months. LLM-agnostic. Cloud-agnostic. Every agent secured by Sinaptic AI Intent Firewall® from day one.
Flexible licensing to scale
15 mins to deploy
zero vendor lock-in
Sinaptic® DROID+ deploys. Sinaptic secures.
Every agent deployed by Sinaptic® DROID+ comes with Sinaptic AI Intent Firewall® built in. You get speed and safety from the same stack.
Sinaptic® DROID+
Runtime Engine
Agent Deployment
LLM Orchestration
Tool Integration
SECURED BY
Sinaptic®
Safety Layer
Sinaptic AI Intent Firewall®
Audit & Compliance
What you get
Built for compliance
Every Sinaptic® DROID+ deployment meets the standards your industry demands.
EU AI Act
GDPR
ISO 42001
ISO 27001
FOR DEVELOPERS
OpenAI-compatible API
Swap your base_url, keep your code. Every call verified by Sinaptic AI Intent Firewall®. Every agent managed by Sinaptic® DROID+.
quickstart.py
from openai import OpenAI
client = OpenAI(
base_url="https://you.droid.plus/v1",
api_key="droid_sk_..."
)
# That's it. Your agent is now secured. OpenAI SDK compatible Sinaptic AI Intent Firewall® built-in 3 lines of code
VERSION HISTORY
DROID+ Release Timeline
Follow the evolution of the platform. Security patches, new LLM models, and system utilities.
v0.5.5 May 27, 2026
Added
- Hot-swappable Custom system prompts — Prompt registry reads overrides from `data/prompts/[agent_name]_system.txt` on request, enabling zero-restart runtime behavioral updates.
- Visual Model Selector & Provider Registry — Live builder dashboard dropdown automatically registers and populates active API key endpoints.
Fixed
- Concurrent SQLite Writes — Implemented a write-lock mutex queue in persistent session storage to prevent database locked errors on parallel agent execution.
- Green Trail carbon estimation — Fixed local conversion multiplier math for Dutch and French cloud regions, preventing overflow during high-load request intervals.
v0.5.4 May 26, 2026
Added
- Visual Builder Upload Support — Visual Agent Builder UI (builder/index.html) now supports attaching files and images in test chat, and displays bubbles for sent images.
- Visual Builder RAG Database — Integrated a RAG Database management card in Preview tab, supporting dynamic lists of indexed files, uploads, and deletions.
- Sandbox File & Image Uploads — Sandbox chat interface (demo/index.html) now fully supports attaching files and images in chat, and lists/deletes documents in the RAG details panel.
- Visual Token Estimation Fallback — Added EstimatePromptTokens to fallback calculate input token usage when the API provider does not return it, explicitly counting model-specific image recognition costs (e.g. 258 tokens for Gemini, 1600 tokens for Claude, 85/255 tokens for OpenAI).
- LLM Regression Tests — Created a comprehensive model regression test suite (regression_llm_test.go) validating direct agent compatibility and execution metrics across all 20+ active 2026 models from OpenAI, Anthropic, Google Gemini, and xAI Grok.
Changed
- Model Pricing Table Updates — Corrected pricing rates in PriceTable for Gemini models (gemini-2.0-flash-lite, gemini-2.5-pro, gemini-2.5-flash) and Grok models (grok-3-fast) to match May 2026 actual pricing.
- Model Registry & Pricing Expanded — Added current 2026 models (gpt-5.4-mini, gpt-5.5, claude-opus-4-7, gemini-3.5-flash, grok-4.3) to PriceTable and GreenTrailTable.
Fixed
- OpenAI Reasoning Models Compatibility — Fixed a request crash on newer reasoning models (e.g. gpt-5-nano, o1, o3-mini) by dynamically mapping the legacy max_tokens parameter to max_completion_tokens, mapping the system role to developer, and omitting the temperature parameter. Also added streaming error propagation.
- Multimodal Block Discarding — Fixed a bug where llm.ChatMessage structures were converted to plain text string payloads before LLM delivery, resulting in the LLM not receiving base64 image data blocks.
- Streaming Chat Memory Persistence — Fixed a bug in openai_compat.go where streaming sessions saved user messages to memory as text-only strings, stripping image blocks for follow-up turns.
v0.5.3 May 25, 2026
Added
- Multi-Channel Chat Gateway Interface — Modular provider-style adapter registry (ChatGateway) to cleanly attach external messaging channels (Telegram, WhatsApp, Viber, etc.) to DROID agents.
- Built-in Reusable Supabase Tool — Added native tool type supabase to execute secure database CRUD actions (select, insert, update) against postgREST APIs using global credentials.
- Prompt Debug Dump Mode — Introduced a debug flag to dump fully rendered messages payloads to data/debug/[agent_name]_latest.json on LLM execution.
v0.5.2 May 25, 2026
Added
- Task Scheduler Engine — Native background scheduling using robfig/cron/v3 to invoke agents or call external webhooks.
- Agent Metadata Schema — Added standard metadata fields (version, author, license, capabilities, a2a_card) to AgentConfig.
- Flexible Keyword Matching — Support for keyword list bypass rules (case-insensitive substring checking) and safe regex pattern matching.
Changed
- Community Agent Limit — Increased the concurrent agent limit from 2 to 20 to allow multi-agent local environments.
v0.5.1 May 18, 2026
Added
- Multimodal (Vision) Support — llm.ChatMessage now supports content blocks (ContentBlocks), enabling base64/URL image processing in Claude/GPT-4o Vision.
- Persistent Session Memory — File-based persistent conversation storage under data/memory/ that survives runtime restarts.
- Dynamic Context Templating — Support for Go text/template syntax in system prompts, populating user details dynamically at runtime.
- Script-based Quick Responses — Regex-based static bypass rules to intercept requests before invoking LLM engines.
v0.5.0 April 10, 2026
Added
- CLI framework — Cobra-based CLI with subcommands: up, down, init, agents, health, version.
- Edition system — Community, Pro, Enterprise editions with feature gating.
- Agent limit enforcement — Community Edition limited to 2 concurrent agents.
- REST API tools — rest_api tool type for calling any HTTP endpoint.
- SinapticAI Community mode — RegExp-based security layer (<2ms latency, ~70% attack coverage).
- Agent Builder UI — visual YAML editor with AI assistant at :8081/builder/.
- MCP and RAG support — Model Context Protocol and file-based RAG with vector search.
v0.4.0 February 28, 2026
Added
- Initial internal release — Basic HTTP server, OpenAI/Anthropic support, YAML loading, and basic SinapticAI security checks.
Ready to deploy your AI agent?
15 mins to deploy. Licensing tailored to your scale. Secured from day one. No AI architects needed.