burnless 0.4.0

Stop burning tokens on repeated context.

Burnless

Intent-compressed intelligence orchestration.

A maestro that orchestrates any LLM from any vendor. Multi-turn agent loops cost O(N²) — Burnless makes them O(N).

Burnless is a vendor-agnostic orchestration layer for multi-agent workflows. You pick the model that conducts the orchestra (Maestro / Brain) — Claude, GPT, Gemini, Mistral, a local Llama, anything — and the models that execute each task (Workers). Tiers are roles, not vendors: gold/silver/bronze/diamond map to whatever CLI you put in config.yaml. Mix providers freely. Run encoder and decoder on a local Ollama model for zero marginal cost on the cheap stages.

On top of that independence, Burnless flips the cost curve. Every turn in a standalone agent loop replays the full conversation as input — token cost on turn N is proportional to N, so total cost across N turns is Θ(N²). Burnless keeps only short capsules in history and shares a cached system prompt across Maestro and Workers. History stays linear; the persistent prefix is billed once per cache window instead of once per turn.

The asymmetry is mechanical, not heuristic. Any provider that charges per input token is subject to the same arithmetic — Anthropic, OpenAI, Google, Mistral, anyone. The reference numbers below use Anthropic's pricing because their cache read/write spread is published and the cheapest to verify ($15/MTok fresh input vs $0.15/MTok cache read — a 100× spread). The mechanism reproduces wherever a provider exposes prompt caching.

Four things, in this order

1. Independence. Any model as Maestro. Any model as Worker. Switch providers in one line.
2. User-enforced rules, not LLM goodwill. You write the routing keywords, the per-tier allowedTools, and the cost budgets in .burnless/config.yaml. With routing.hardcore_filter: true (or BURNLESS_HARDCORE=1), the Maestro cannot escape to a higher tier than the keyword router resolved — no quiet upgrades to Opus for tasks the rules said belong to Haiku. allowedTools is enforced by the worker CLI itself, not hinted at in the prompt: when bronze ships with Read,Bash, it physically cannot Edit. Bypass requires an explicit --force from the human.
3. Three compression layers. Deterministic minifier (regex, zero cost), semantic encoder (small model, ~$0.001/turn), optional LLMLingua-2 (CPU-only, no API). Each layer is independent and additive.
4. Math, not marketing. 88% cheaper at turn 10 by arithmetic on the published pricing pages. Verify with python bench/run.py --turns 8 and your own API key.

The numbers

Two views, both reproducible on your machine.

Real API run — 10 turns against claude-opus-4-7, 23k-token prefix, no mocks, raw response.usage (actual spend: $5.76):

Scenario	Cost	vs no-cache
A — Standalone, no cache	$4.66	—
B — Standalone + cache	$0.65	−86.0%
C — Burnless Maestro	$0.45	−90.3%

Reproduce: ANTHROPIC_API_KEY=... python bench/run.py --turns 10 (~$6).

Monte Carlo simulation — 30 runs × 100 turns × 4 scenarios. Per-turn input/output sampled Uniform(2k, 10k) / Uniform(200, 1500), capsule compression Uniform(0.20, 0.30). Zero API cost:

Scenario	Mean	vs Pure Opus
A1 — Pure Opus 100	$532.61	—
A2 — Pure Sonnet 100	$105.42	−80.2% (5× cheaper)
B — Free-pick (Opus/Sonnet)	$328.74	−38.3% (1.6× cheaper)
Z — Burnless	$33.35	−93.7% (16× cheaper)

The interesting row is B. A developer alternating Opus and Sonnet ad-hoc — what most people actually do — costs 3× more than just sticking with Sonnet, because every model switch invalidates the prefix cache. Burnless is 10× cheaper than B and 3× cheaper than the disciplined "all Sonnet" strategy, because Brain stays fixed (cache hot) while workers tier down to Haiku where they can.

Reproduce: python bench/v2.py --runs 30 --turns 100 --seed 42. Zero cost, no key.

For the formal derivation — including why Burnless only loses at N = 1 — read MATH.md.

Design decisions

The 88% number is an outcome. These are the calls that produced it, in the order they were made.

1. Treat the cost curve as math, not engineering. Multi-turn agents replay full history every turn. Tokens billed across N turns sum to Θ(N²) — that is arithmetic on the pricing page, not a property of any SDK. Once the problem is stated as O(N²), the only useful question is what to truncate. Everything else follows.

2. Brain stores capsules, not transcripts. The Brain's conversation history holds ~80-char single-line summaries of each prior turn, not the raw exchange. Full output stays on disk, read on demand. This is the single change that flips the curve to O(N) — every other layer compounds on top of an already-linear baseline.

3. Shared prefix cache across models. If two models from the same provider see a byte-identical system prompt with cache_control set, they hit the same prefix cache. Switching Opus → Sonnet mid-session does not invalidate it. Brain and Worker can be different models and still amortize the 23k-token system prompt at read price ($0.15/MTok) instead of write price ($15/MTok). The 100× spread is the lever.

4. Tiers are roles, not models. gold/silver/bronze/diamond map to commands in config.yaml, not to Opus/Sonnet/Haiku. Any model can be Brain. Any model can be Worker. GPT-4o as Brain delegating to Codex workers is a one-line config change. Hardcoding tier→model would have made the orchestration layer a single-vendor wrapper instead of a pattern.

5. Determinism before LLMs. Layer 1 of the compression stack is pure regex plus a glossary — no model call, zero latency, zero cost. Filler phrases, normalized whitespace, abbreviations applied before the encoder ever sees the text. A semantic compressor running on already-clean input is cheaper and more stable than one fighting prose. Cheap stages run first for a reason.

6. Pluggable encoder. The semantic compression layer accepts Haiku, LLMLingua-2 (Microsoft Research, XLM-RoBERTa, CPU-only), or any local model — selected per-invocation via --encoder. Coupling the architecture to one specific compressor would have tied savings to one model's pricing. Keeping each layer swappable means the stack improves automatically as local models get better, with no API change.

7. The benchmark is the proof. bench/run.py is short, dependency-light, hits the Anthropic SDK directly with no mocks, and writes raw response.usage to JSON. Anyone can rerun it, contest the numbers, and open an issue with their own results file. We did not write a marketing page about savings; we wrote a script that produces them and invited disagreement. That is the only honest way to publish a cost claim.

Install

pip install burnless
burnless setup               # one-time, detects local agents and keys
burnless init                # inside any project directory

Python 3.10+. Tiers map to whatever models you configure — mix providers freely.

Or install from source:

git clone https://github.com/rudekwydra/burnless.git
cd burnless && pip install -e .

Any model. Any role. Full control.

Tiers are roles, not models. You decide what runs each role — and any model can be the Brain.

# .burnless/config.yaml — example: GPT-4o as Brain, Sonnet as executor, Codex for code
agents:
  gold:    { command: "openai api chat.completions.create -m gpt-4o" }
  silver:  { command: "claude --model claude-sonnet-4-6 -p --allowedTools Read,Edit,Write,Bash" }
  bronze:  { command: "claude --model claude-haiku-4-5 -p --allowedTools Read,Bash" }
  diamond: { command: "codex exec --sandbox workspace-write" }

Gemini as Brain, DeepSeek for execution (both have published cache pricing):

agents:
  gold:    { command: "gemini -m gemini-2.0-flash-thinking -p" }   # Brain
  silver:  { command: "deepseek chat --model deepseek-chat -p" }    # execution
  bronze:  { command: "ollama run qwen2.5-coder" }                  # local, zero cost

Or Sonnet as Brain delegating to Codex workers:

agents:
  gold:    { command: "claude --model claude-sonnet-4-6 -p" }   # Brain
  diamond: { command: "codex exec --sandbox workspace-write" }   # code execution
  bronze:  { command: "ollama run llama3" }                      # local model, cheap tasks

Each tier gets its own allowedTools, routing keywords, and cost budget. The routing layer reads your task description and picks the right tier automatically — or you override it explicitly.

The O(N²) → O(N) math applies to any provider that charges per input token. Burnless is the orchestration and caching layer, not a wrapper for one API.

Taking it further: the encoder and decoder — the models that compress user messages into capsules and expand capsules back into natural language — can run on a local model at zero marginal cost:

agents:
  bronze: { command: "ollama run llama3.2" }   # capsule encoder/decoder — $0
  silver: { command: "claude --model claude-haiku-4-5 -p" }
  gold:   { command: "claude --model claude-sonnet-4-6 -p" }   # Brain
  diamond: { command: "codex exec --sandbox workspace-write" }

As local models improve, more tiers move to zero cost. The expensive models (Opus, GPT-4o, Gemini Pro) handle only what requires genuine reasoning — and they do it with a cached prefix and a linear history.

The O(N²) → O(N) math applies to any provider that exposes prompt caching: Anthropic, OpenAI, Google Gemini, DeepSeek, Mistral, Qwen — and any local provider via Ollama. If the provider charges per input token and supports a prefix cache, Burnless works.

Three compression layers

Each layer is independent and additive:

Layer	What it does	Cost	When it fires
1. Deterministic minifier	Strips filler phrases, normalizes whitespace, applies glossary abbreviations	Zero — pure regex	Every turn, before encoder
2. Semantic encoder	A small model (Haiku, GPT-4o-mini, local Llama) compresses prose into structured capsule format	~$0.001/turn	Every turn
3. LLMLingua-2 (optional)	XLM-RoBERTa token classification (Microsoft Research, GPT-4 distilled)	Local CPU, no API	Long inputs, --encoder llmlingua2

The 88% cost reduction in the benchmark comes primarily from Layer 3 of the architecture — shared prefix cache + linear capsule history. Layers 1 and 2 compound on top of that.

Compression modes

Three modes control the cost × epistemic fidelity trade-off — how much of the argumentative trajectory a session preserves:

Mode	Layers active	Anchor preserved	Friendly output	Savings	Use when
light	Minifier only (L1)	Yes	On	~40%	Architecture debates, decisions that may need revisiting
balanced (default)	Minifier + encoder (L1+L2)	No	On	~88%	Project execution, multi-step implementation
extreme	All layers (L1+L2+L3)	No	Off	~93%+	CI/CD pipelines, batch automation, no human in the loop

Anchor preserved means the Brain's capsules retain enough argumentative structure that prior decisions remain revisable — you can genuinely reconsider, not just append. balanced discards the trajectory and keeps only the semantic result: the Brain knows what was decided, not why. Workers are always epistemically pure regardless of mode — they receive a clean task without the Brain's debate history.

compression:
  mode: light   # light | balanced | extreme

Or per-invocation: burnless --mode light "review this architecture decision".

The formal derivation of why capsule compression reduces both cost and anchoring bias is in MATH.md §10.

How it works

Brain. A thin orchestrator — any model you configure — that holds the plan, decides what to delegate, and reasons over results. Its conversation history contains only capsules — single-line summaries of past turns, ~80 characters each.

Worker. A delegated execution (any tier, any provider — local Ollama, Codex, Claude, GPT, Gemini) that receives one task, the cached system prompt, and the relevant capsules. It runs, returns a compact result, and exits. Raw output is written to .burnless/logs/dNNN.log, never replayed into the Brain.

Capsule. The compact handoff between turns. The Brain reads the capsule; the full log stays on disk and is read on demand. This is what flips the cost curve from quadratic to linear.

Shared cache, kept hot by architecture. Brain and Worker use a byte-identical persistent prefix marked with the provider's prompt-caching directive (Anthropic: cache_control: {"type": "ephemeral", "ttl": "1h"} — 1h, not the 5min default). The session is append-only on disk (.burnless/maestro_session.jsonl): every turn extends the message array without rewriting earlier blocks, so the cached prefix stays bit-identical and lookups hit. Up to 4 nested cache breakpoints (Anthropic's per-request limit) sit at system → memory → plan → capsules, shared by Brain and all Workers. When capsules accumulate (~30+), Haiku pre/post-compacts them into a denser prefix block which is then re-marked — the cached region stays bounded and stays hot indefinitely within the TTL.

The one known gap: if a session sits idle > 1h with zero calls, the TTL expires and the next call pays write price again. A --keepalive mode (1-token ping every ~50min for daemon-style usage) is on the v0.4 roadmap; not in v0.3. See MATH.md §8 for the full derivation of why the cache_read assumption is load-bearing for the O(N) result.

Benchmark

The benchmark in bench/run.py is the source of truth for the table above. Three scenarios run through a real provider SDK directly with no mocks; costs come from response.usage exactly. The reference run uses Anthropic because their cache pricing is published and easiest to reproduce — adapters for OpenAI and Gemini are tracked in the issues.

• A — standalone, no cache, full history each turn
• B — standalone, system prompt cached, full history each turn
• C — Burnless Maestro: cached system prompt + capsule history

Reproduce the math without an API key:

python bench/run.py --project 50

Reproduce empirically (real API calls, ~$5 for 8 turns):

ANTHROPIC_API_KEY=sk-ant-... python bench/run.py --turns 8

Raw results land in bench/results/run_<timestamp>.json for inspection.

CLI

burnless                     # interactive shell (Brain)
burnless plan "<objective>"  # write a plan to .burnless/maestro.md
burnless delegate "<task>"   # create a delegation, route to a tier
burnless run d001            # execute it (worker streams to live panel)
burnless status              # current plan + open delegations
burnless metrics             # token counter + audit ledger

State lives entirely under .burnless/ in your project. No hosted backend.

vs. LangChain / CrewAI / AutoGen

Burnless is not a competing orchestration framework — it is an optimization layer that sits under your existing agent logic. The distinction matters:

	LangChain / CrewAI / AutoGen	Burnless
Primary focus	Agent connectivity and orchestration	Cost reduction and cache efficiency
Memory model	Sliding window or RAG	Compact capsules, Brain-led
Cost shape	Θ(N²) — grows quadratically	Θ(N) — grows linearly
Dependencies	Heavy libraries, many abstractions	Lightweight CLI (pip install burnless)
Hosting	Local or cloud	100% self-hosted — zero data retention
Provider lock-in	Varies	None — any CLI, any provider, any model

You can wrap a LangChain agent as a Worker. The Brain→Worker pattern is compatible with any existing agent framework — Burnless manages the context budget and cache strategy; your agent handles the task logic.

When Burnless is not the right tool: single-turn queries (N = 1), one-off scripts with no repeated context, or workflows where a managed cloud platform is the explicit requirement (in that case: waitlist for Burnless Cloud at burnless.pro).

Contributing

Issues, PRs, and benchmark contestation are all welcome. The benchmark script is intentionally short and dependency-light so you can read it end-to-end and disagree with concrete numbers. If your workload produces a different ratio, open an issue with the JSON from bench/results/ — that is exactly the conversation worth having.

Status — what works today, what's roadmap

The architecture is provider-agnostic by design. Current implementation status:

• ✅ Workers: shell out to any CLI (claude, codex, openai, gemini, ollama, anything). Configure per tier in config.yaml. Works today.
• ✅ Routing, capsules, exec_log, three compression layers, shared system prompt: provider-neutral, work today.
• ✅ Reference benchmark: uses Anthropic SDK because their cache pricing is published and easiest to reproduce. The math reproduces wherever a provider exposes prompt caching.
• ⚠️ burnless brain interactive command: uses the Anthropic SDK in-process today. OpenAI, Gemini, and OpenRouter adapters are tracked in v0.4. If you want to skip the in-process Brain, burnless delegate + burnless run already cover the full Brain→Worker loop using whatever CLI you configured.
• ✅ PyPI release: pip install burnless — version 0.3.0 live at https://pypi.org/project/burnless/.

Honest about gaps. PRs welcome — especially for the OpenAI/Gemini Brain adapter.

License

MIT. See LICENSE.

---

Repo: github.com/rudekwydra/burnless