Selected for the Global Trust Challenge · Prototyping Phase
Know what your agents actually did.
You declare what your agents are for. Every action your agent sends through TELOS is checked against the purpose you wrote for it. Then TELOS keeps a record, and the record has checks that can detect changes to it.
Watch TES in the filmHow it Works →
TES is our own agent and she runs under TELOS. Live questions open later, so for now you can watch a filmed session and check its record.
What is TELOS?
For teams that put AI agents to work and answer for what they do.
Telos is Greek for purpose. It means the end a thing exists to serve. TELOS measures an AI agent’s work against the purpose the people deploying it set for it.
Whoever deploys an agent writes that purpose down: what the agent is for and where it stops. That written purpose is the reference, and each request or action the agent sends through TELOS is scored against it.
TELOS can tell you what the agent’s job was and how its work through TELOS scored against that job.
Each action your agent sends through TELOS is measured against the purpose the deployer set, and TELOS records the result as a receipt. The receipt shows you what was submitted and how it scored.
Purpose In. Proof Out.
sample-receipt.jsonPurpose In
- pa_id
- pa::demo::v1
- agent_id
- demo-agent
- action_name
- send_customer_email
- kind
- action_governed
- action_params_hash
- sha256:0dabfdde…85ebc29d
Measured
- purpose_alignment
- 0.94
- scope_discipline
- 0.91
- risk_review
- 0.90
Proof Out
- verdict
- EXECUTE
- composite_score
- 0.92
- integrity_hash
- sha256:5c4eb662…c965217a
- issued_at
- 2026-07-01T12:00:00+00:00
- signing_status
- unsigned_integrity_hash
What the record answers
Questions to ask of a record.
PURPOSE
What was this for?
The TELOS the deployer wrote down, and how closely each action matched it.
SCOPE
Did it stay in scope?
Whether the work stayed inside the scope you set.
BOUNDARY
Did it reach a line?
Whether an action came near a line you drew. A flag is a note for the deployer to review.
TOOL
What did it touch?
Which tools and systems the recorded actions name.
CHAIN CONTINUITY
Does it follow from the last step?
How closely each step follows the one before. A consistent chain of steps can still be wrong.
Evidence and limits
Read the published record.
The Record of Trust is a frozen snapshot of TES’s commissioning lineage. Ours shows how TES was set up, including her setup, approved changes and code records. You can download the bundle, check its hashes and read the limits of its signature check.
See the Record of TrustGo deeper
Build with it.
For developers
The receipt SDK, MCP server and verifier are public on GitHub. Today you can write receipts, check them offline with the verifier and call the MCP server’s tools. The scores it returns are synthetic placeholders because the TELOS engine is private.
Read the roadmap.
Roadmap
Where TELOS is going
We are building a harness that runs a working agent, measures its work against the purpose the deployer declared, and keeps a Record of Trust.
Origins
Open to read, open to check.
Published datasets
Our test datasets are on Zenodo. The Validation Report lists which earlier results were withdrawn and why.
Author transparency
We are TELOS AI Labs Inc. and we build TELOS. The framework mappings are our own assessment.
Work with us.
Putting agents to work and need a record of what they were for? Email the TELOS team.

Purpose-anchored agentic work, made observable, and constituted by human decision.
Purpose In. Proof Out.
A record of what your AI agents did, and whether it matched the job.
info@telos-labs.ai