The agent and the knowledge base, working as one. — SYMBIOS AGENT is the first engineering agent built around a live World Model — a structured internal representation of your system's topology, history, and risk — running autonomously, around the clock. Built for ML and infra teams with complex, dependency-heavy systems.
This is not a general-purpose AI assistant. It's infrastructure.
No ticket. No wake-up call. Here's what happens next.
Most engineering tools wait to be asked. SYMBIOS AGENT discovers work, assesses risk, and executes — on its own schedule, without a human in the loop.
A persistent process running three independent passes: health checks, task discovery, and self-directed planning. Restarts automatically on failure.
A live internal representation of your system's structure. The agent queries it before any action to predict consequences and decide whether to proceed.
A structured, typed, persistent store. Every run writes back to it. The agent's ground truth for what your system is, what happened, and what was decided.
Before executing any task, the agent traverses the dependency graph, checks incident history, and issues a go / hold / abort decision — without being asked.
Risk is computed via topological traversal of the full dependency graph — not just immediate dependents.
Evidence freshness threshold: 24h. Returns graceful fallback if stale. Confidence: 0.0–1.0.
Not a conversation log. A typed, indexed record of your system's modules, incidents, evaluations, test history, and the agent's own reasoning — all queryable, all versioned.
Most tools that claim memory store what you told them in chat. SYMBIOS AGENT indexes what your system is — its topology, its failure patterns, its evaluation trajectory — and uses that to decide what to do next, without asking.
The KB is the agent's ground truth. Every health check, every task completion, every self-generated plan writes back into it. Evidence freshness is tracked. Confidence degrades with staleness. The agent won't act on stale information.
Drop a task file in kb/pending_tasks/ and the agent picks it up within 300 seconds — checks the world model for blockers, assesses cascade risk, and starts work. No conversation required.
Running a local model isn't just swapping an API endpoint. Instruction-tuned models have different sensitivity to system prompt structure, verbosity, and framing. SYMBIOS AGENT ships a dedicated prompt policy engine that resolves the right profile per model, per intent — automatically.
Each profile shapes system prompt structure, verbosity, and instruction framing to match the task type and model behavior.
Before routing to a local model, the request is classified into a lane. The lane determines which model and which prompt profile are applied.
Every local model call goes through a centralized policy resolver that determines the effective profile, mode, and routed model — with full audit trail.
local_prompt_profile_hintQuery the dependency graph, assess cascade risk before a deploy, read incident history, control the daemon — all from your existing toolchain.
| Method | Endpoint | Description |
|---|---|---|
| GET | /api/v1/world/health | Service up check |
| GET | /api/v1/world/state | Full project snapshot |
| GET | /api/v1/world/dependencies/{module_id} | Reverse dependency graph for a module |
| GET | /api/v1/world/cascade?modules=a,b,c | Cascade risk for a set of modules |
| GET | /api/v1/world/incidents/{module_id}?days=7 | Incident history, 7-day default window |
| GET | /api/v1/world/loop/status | Daemon state: running / stopped / failed |
| POST | /api/v1/world/loop/start | Start daemon (async) |
| POST | /api/v1/world/loop/stop | Graceful stop |
| GET | /api/v1/world/loop/metrics | All loop counters — tasks, incidents, plans, uptime |
| GET | /api/v1/world/plans | Self-generated work queue |
No other autonomous agent ships with native GPU and ML infrastructure skills. Deploy inference servers, switch GPU modes, optimize cache — without manual orchestration.
The capabilities below are unique to SYMBIOS AGENT. Reactive agents — IDE assistants, cloud coding agents, memory bolt-ons — share none of them in combination. To our knowledge at time of publication, based on publicly documented architectures and capabilities.
| Capability | SYMBIOS AGENT | Memory tools | IDE agents | Cloud agents |
|---|---|---|---|---|
| Live dependency graph (World Model) | ✓ | — | — | — |
| Cascade risk before every action | ✓ | — | — | — |
| Autonomous 3-pass loop (no trigger) | ✓ | — | — | — |
| Self-generated work proposals | ✓ | — | — | — |
| Native GPU / ML infra skills | ✓ | — | — | partial |
| Persistent structured KB | ✓ | conversation only | static files | — |
| JSONL audit trail + reasoning logs | ✓ | — | — | — |
| Self-hosted, $0 inference cost | ✓ | self-host option | — | enterprise tier |
| Watchdog + exponential back-off restart | ✓ | — | — | — |
| REST API for external toolchain | ✓ | ✓ | — | ✓ |
Run on your own hardware with Ollama or vLLM — $0 per token. You pay for the system, not the usage. Infrastructure and energy costs remain yours.
The system runs entirely on your infrastructure. We don't access your environment. Support is async — documentation, configuration guidance, and custom feature development delivered remotely.
Team and Studio tiers are in early access. Tell us about your use case — we onboard in order of fit and readiness.
SYMBIOS AGENT is in limited release. We're onboarding engineers and teams ready to run a system that doesn't wait to be asked.
Request Access