Lossless token reduction for LLM developers.
Athena sits in front of your LLM calls and rewrites bulky JSON, logs and tool output into smaller equivalents before your provider sees them. Every value is checked present. You bring your own provider key. Savings are measured on your content, never promised.
See your own saving first
Paste a sample of a tool result, API response or log. Athena runs the lossless pipeline and reports the real saving. Up to 100,000 characters. Processed in memory, never stored. Table encodings (JSON, logs, markdown tables) are included in every paid plan.
Where Athena saves the most
Measured by Oversight Labs on its own samples with the cl100k_base tokenizer. Results vary with your content and provider, so try the measure tool on your own data.
Small changes, verified.
Athena only keeps a smaller form when it can prove nothing was lost.
Send your messages
Post your chat messages or tool output to the compress endpoint, or use the OpenAI compatible proxy with your own provider key.
Athena shrinks what is safe to shrink
Whitespace and markup are cleaned, repeated lines and exact duplicate blocks are collapsed, and JSON, timestamped logs and markdown tables are re-encoded into compact forms.
Every value is checked
JSON must keep every key and value, logs must decode back to the exact original, and anything that fails a check is sent unchanged.
You get the result and the numbers
The response includes tokens in, tokens out and the ratio for that request, and your usage page keeps the running ledger.
Lossless and verified
JSON keeps every key and value, logs decode back exactly, and failed checks fall back to your original text.
Works with prompt caching
Output is deterministic and prefix stable: adding messages never changes earlier ones, so provider prompt caches keep hitting. We tested the stability; we have not measured cache savings.
Nothing stored
Message content is processed in memory. Your provider key is used once per request and never stored. API keys are stored only as hashes.
Usage ledger
Tokens in, tokens out and ratio for every request, kept per account and shown on your usage page.
Two calls
curl https://oversightlab.ai/athena/v1/compress \
-H "Authorization: Bearer ath_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"messages":[{"role":"tool","content":"...your tool output..."}]}'The response returns the smaller messages and a metrics object. The proxy at /athena/v1/chat/completions is OpenAI compatible: send your provider key in X-Provider-Key and name the provider in X-Provider (openai, deepseek, openrouter or gemini). Your provider key is used for that request only and never stored. Streaming is not supported yet. Usage: GET /athena/v1/usage.
Five plans. Pay monthly, cancel any time.
Prices in pounds sterling. Oversight Labs is not VAT registered, so no VAT is added. You will see our Terms and Privacy Policy before paying.
Pro
- Whitespace and markup cleanup, repeat collapse, exact duplicate removal
- JSON minify with a value identical check
- Table encoding for JSON, logs and markdown tables on up to 5,000 messages a month
- 25M input tokens a month
- 60 requests a minute
- Usage ledger and free measure tool
Team
- Everything in Pro
- Table encoding on every message with no monthly cap, up to 30 table compressions a minute
- 100M input tokens per seat a month
- 120 requests a minute
- 1 to 50 seats
Team Plus
- Everything in Team
- No limit on table encoding
- 250M input tokens per seat a month
- 240 requests a minute
- 1 agent key per seat for your automated workloads
Team Max
- Everything in Team Plus
- Athena MCP server for your AI tools: athena_compress, athena_measure and athena_usage
- 250M input tokens per seat a month
- 240 requests a minute
Studio
- Everything in Team Max
- 500M input tokens per seat a month
- 480 requests a minute, our highest limit
- 3 agent keys per seat
- 3 Oversight Crew Agents included (coming soon, not yet available)
Athena Enterprise
For larger teams and custom arrangements: volume pricing, contract terms and support are agreed directly. Book a call.
Billing is monthly through Stripe and you can cancel from the billing portal at any time. Allowances are per month. If you are a consumer you have a 14 day cancellation right unless you ask us to start the service immediately; see the Terms.