Evidence
Sample receiptSigned demoArchitectureResearch and limitsPublications
How it works

See what your agents did, what purpose applied, and when they needed a human.

As AI agents move from answering questions to doing real work, organizations need to know what happened. TELOS shows what an agent did, what assigned purpose applied, what the resulting record shows, when it escalated, and what evidence remains afterward.

The gap

Agents are doing real work. The answers lag behind.

Approvals do not scale to the volume of consequential actions. Logs record calls, not whether the work matched its purpose. TELOS sits alongside the agent and keeps the record a human can check.

The workflow

Six steps, start to finish.

TELOS does not run your agents. It makes the work they do observable against the specification a human gave them, and it delivers tamper-evident evidence of what the agent did, when it did it, and how.

01

A human specifies the work

Purpose, escalation conditions, and evidence expectations. The specification is the human's, in writing, before the agent moves.

02

The agent performs the work

Inside the declared context, on whatever stack you already run. TELOS is not a model and not an agent framework.

03

TELOS observes runtime action

On instrumented paths, consequential actions are recorded against the specification: the action, the purpose it was measured against, per-dimension scores, and a review verdict.

04

Human review signals are surfaced

Consequential calls become visible to your people, and the human decision lands in the record.

05

The evidence preserves what happened

On instrumented paths, observed actions produce durable, tamper-evident evidence: records with an integrity hash anyone can recheck offline, and no raw arguments echoed.

06

Teams review purpose discipline

The accumulated evidence supports human review of how observed actions compared with assigned purpose over time. The work leaves evidence.

Who it serves

Everyone ends up asking the same thing.

Operators

Teams deploying agents

Agency owners, RevOps leads, SMB operators. You want to see what the agent did and show a human stayed in authority, without clicking approve on everything.

Builders

People instrumenting workflows

Developers and automation consultants. You want evidence and observability added to real agentic workflows without rebuilding the stack. The SDK is free for exactly this.

Readiness

Compliance and risk stakeholders

Advisors, risk managers, insurers. You want an evidence surface to exist before consequential AI workflows become a liability question. That evidence is the surface.

Status · honest

What works today, what is coming.

The historical July 2026 signed capsule and offline bundle are available. Verification checks the recorded evidence, not a new benchmark run. The public standalone verifier checks unsigned receipt payload hashes, a different contract.

Wired today
  • Runtime observation against a declared purpose on instrumented paths
  • Per-action records with an integrity hash anyone can recheck offline
  • Human-review signals recorded into the evidence
  • The free telos-receipt 0.1.0 SDK builds unsigned receipts. The telos-mcp 0.1.1 server provides synthetic stub interfaces, with no live governance engine wired.
Coming
  • Signed receipts (records are integrity-hashed and unsigned today)
  • Broader coverage beyond instrumented paths

Purpose In. Proof Out.

TELOS makes AI-agent work visible and human-accountable.

info@telos-labs.ai