An intelligent orchestration layer for LLMs.
Devtor optimizes cost, performance, and reliability by dynamically managing prompts, models, context, and execution strategies — continuously improving from past interactions.
A control plane between you and your LLMs
Devtor orchestrates prompts, models, context, and execution — optimizing cost, performance, and reliability. It picks how, when, and which model to run, learns from every session, and works via CLI, Locally Hosted Webportal, or IDE extensions.
The right model, context, and execution strategy for every task.
- ✓Hybrid super-agent + cheap sub-agents
- ✓Granular plans · pre-run cost estimates
- ✓Smart context · latest prompt only
- ✓Cross-user learning · shared fixes
- ✓CLI + Locally Hosted Webportal + IDE extensions
Prompt optimization
Tune prompts per model tier. Cheaper models, production output.
Smart context
Relevant messages and files only. No full-history replay.
Cost control
Estimates, budgets, smart routing. Power models when needed.
Organizational learning
Resolved failures indexed org-wide. Same problem, instant fix.
Your Infrastructure. Your Models.
Our Agent Orchestration.
Stop being locked into single AI vendors. Devtor is fully model-agnostic, giving you the power to mix local offline intelligence, specialized high-speed endpoints, and private enterprise deployments inside a unified routing layer.
Zero Provider Lock-In
Swap providers with a single config change. Connect commercial APIs, local nodes, or private instances.
Secure Local Workloads
Run offline tasks using Ollama, Llama.cpp, or vLLM. No external network request is ever sent for local workflows.
Unified Configuration Spec
A single standardized spec handles diverse model prompt templates, token limits, and fallback strategies behind the scenes.
Failover Resilience
Configure fallback lists. If your main model hits rate limits or experiences downtime, traffic transparently switches to your standby model.
Hybrid Local & Cloud
Most PopularRun local open-source models for draft generation and security-critical files, and scale to premium cloud LLMs only when deep reasoning is needed.
Claude 3.5 Sonnet
Anthropic API
Llama 3.1 8B
Ollama (Local)
DeepSeek-V3
DeepSeek API
Flow: Drafting, tests & simple refactors run 100% locally on your machine at zero cost. Multi-file plans automatically upgrade to Claude 3.5 Sonnet.
# ~/.config/devtor/models.yaml
providers:
anthropic:
api_key: env.ANTHROPIC_API_KEY
models:
- claude-3-5-sonnet-latest:
tier: advanced
routing: "reasoning, planning"
ollama:
host: "http://localhost:11434"
models:
- llama3.1:8b:
tier: local
routing: "drafting, refactoring, tests"Orchestration in three phases
Plan, route, execute — Devtor picks the right model and learns every run.
Plan First
Analyze before you infer.
Analyzes task complexity, token usage, and execution path before any model runs.
- Granular plans when complexity demands them
- Pre-execution token and cost estimates
- Token budgets enforced before compute starts
Route Intelligently
Right model. Right price.
Picks the cheapest capable model. Prompt tuning and hybrid sub-agents keep quality up, cost down.
- Smart model selection per task type
- Prompt optimization for cheaper model tiers
- Hybrid orchestrator + sub-agent delegation
Execute & Learn
Run once. Route smarter next time.
Runs the plan, tracks spend, caches patterns. Think mode lets you steer mid-run.
- Interactive think mode for mid-run corrections
- CLI + local web IDE for flexible workflows
- Smart context — only relevant messages per call
Built for engineering teams
who take AI seriously.
16 capabilities across five areas — routing, planning, context, execution, and learning.
Model routing
Pick, tune, and delegate across model tiers.
- Prompt Optimization LayerTune prompts so cheaper models match production quality.
- Hybrid Model ArchitectureExpensive orchestrator delegates to cheap sub-agents. No wasted spend on simple tasks.
- Smart Model SelectionAuto-picks the best model per prompt, using expensive models only when needed.
- Model Limitation AwarenessKnows model limits — escalates or reroutes instead of hallucinating.
Planning & cost
Plan before you spend. Stay inside budget.
- Granular Plan ModeGenerates detailed step-by-step plans when needed, and skips planning for simple tasks.
- Pre-Execution Token EstimationSee token usage and projected cost before build commands run.
- Token Budgeting SystemSet token budgets per session, task, or team — enforced before compute runs.
- Structured Plan OutputOutputs interactive plan.html — reviewable decisions, not markdown walls.
Context & memory
Send only what matters. Control what persists.
Execution & workflow
Run, steer, build skills, collaborate.
- Interactive Think ModePause, redirect, or correct the agent mid-run — no full restart.
- Skill BuilderBuild skills with test and evaluation modes to benchmark and fine-tune behavior.
- CLI + Local Web IDEProvides a terminal CLI and localhost web IDE with embedded chat.
- AI Collaboration ModeShared agent sessions and handoffs — break out of solo AI silos.
Team learning
Shared fixes and repo-aware setup.
CLI, Webportal & IDE Extensions. Ready when you are.
Terminal CLI, Locally Hosted Webportal, and IDE extensions. Switch between interfaces with your existing provider keys.
Install the CLI
yarn, npm, or homebrew
Add provider API keys
Reads from your .env
Run devtor init
Auto-detects your codebase
Run your first task
Natural language in terminal
- Node.js 18+ · one LLM provider key
- CLI, Webportal & IDE extensions
- Pre-run estimates · token budgets