# Navige (slim)

> Navige is a hosted gateway between AI agents and the model APIs and MCP tools they call. It prices each call per business job, enforces spend caps, policies and approvals, and can cache or reroute calls to cut cost.

- Full: https://www.anchorterminal.com/tools/navige.md (~10,900 tokens) · this version ~2,580 tokens · JSON https://www.anchorterminal.com/tools/navige.json · canonical https://www.anchorterminal.com/tools/navige
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-18 · pricing https://www.anchorterminal.com/tools/navige#pricing · reviews https://www.anchorterminal.com/tools/navige#reviews
- Licence: CC BY 4.0, credit Anchor Terminal (anchorterminal.com) and link this page when you pass anything on

**E · 42.8/100 · rank #1012 of 1133 · #16 in Agent observability & evals · not agent-ready · confidence medium**


Assessment: Setup is a base-URL change, keys are per agent and hashed, and the docs state failure behaviour per action class. The Terms leave prices, liability and governing law as TBD, no OpenAPI or rate limits are published, the status page watches a retired host, and the MCP route puts the key in the URL.

## Facts

- Kind: HTTP API · vendor: Navige Ltd · category: Agent observability & evals · legal entity: Navige Ltd · provenance 65/100
- Endpoint: `https://api.navige.ai` (HTTP)
- Auth: API key · pricing: Freemium · x402: no · licence: Proprietary hosted service under Navige's Terms of Use (version 1.0, effective 13 October 2026) and Data Processing Agreement (version 1.2, effective 14 October 2026). The `navige` npm SDK and the `n8n-nodes-navige` node are MIT
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The hosted gateway at https://api.navige.ai (model proxy, `POST /enforce`, `POST /mcp`, `POST /usage`, `POST /v1/transactions/:id/outcome` and the keyless receipt, verification and health endpoints), read from the docs, onboarding guide, pricing, legal pages and unauthenticated requests on 17 October 2026. No governed call was made
- Release history: Published as TrustLoop, with npm `trustloop` 1.0.0 on 7 March 2026. Renamed Navige with npm `navige` 2.0.0 on 26 July 2026 (navige.ai registered 25 July 2026), 2.0.1 on 12 August and 3.0.0 on 7 September. Product Hunt launch in mid-October 2026, per the badge on the home page
- Routes: Model proxy for OpenAI, Anthropic, Google (alias gemini), xAI (alias grok) and Bedrock, which Navige re-signs with an assumed role in the customer's AWS account. MCP endpoint that forwards tool calls to MCP servers registered in the console. Decision API for actions outside a proxy. Usage import for spend Navige did not see
- Decision API: `tool` (required, up to 200 characters), `system`, `classification` (`read`, `write`, `destructive`, `financial`, `unknown`), `arguments`, `transaction`, `workflow` and `wait` (default true, holds up to five minutes). Arguments are stored masked and with secrets removed. The decision is advisory and the caller acts on it
- Plans: Watchtower, free, unlimited agents, observes and records policy decisions as advisory, with kill switch and spend caps. Paid, $9 a month or 20 per cent of measured savings, whichever is greater, with enforcing policies, approvals and the cost levers. Enterprise by contact
- Cost levers: Duplicate suppression for reads, result capping, tool scoping, context trimming, Anthropic prompt caching (on by default for new organisations), output ceiling and model routing within the same provider, never above the model asked for. All off until switched on, apart from prompt caching
- Correlation: `x-navige-workflow` for the kind of job, `x-navige-transaction` for one run (makes retries count once), or a W3C `traceparent`. Each bounded at 200 characters and rejected if longer. Claude Code sessions correlate automatically. Outcomes are reported with `POST /v1/transactions/:id/outcome`
- Errors: 400 malformed, 401 with `missing`, `not_found`, `revoked` or `malformed`, 403 `agent_inactive`, `agent_expired` or kill switch active, 404 unknown provider, 429 blocking budget reached, 503 Navige unavailable (`decision: "error", unavailable: true` on `/enforce`). Degraded decisions carry one of five cause codes. Provider errors pass through unaltered
- Failure behaviour: When policy, kill switch, budget or counter state cannot be read, `read` and `unknown` calls fail open and `write`, `destructive` and `financial` calls fail closed, configurable per class. A recording failure after a successful upstream call still returns the response
- Rate limits: No numbers published in the docs, onboarding guide or terms
- Data handling: Per DPA 1.2, prompt and completion text through the proxy is not stored, tool call arguments are stored masked, no automatic time-based deletion runs, data is held by Railway in the United States, and no data trains models. Sub-processors Railway, Resend, Stripe and Framer
- Verification: Keyless `GET /api/receipts/:actionId` for signed decision receipts, `GET /.well-known/navige-signing-key` (an Ed25519 public key, fetched on 17 October 2026), batch and record verification against Polygon anchors (optional, off by default), and an organisation badge
- SDKs: `navige` 3.0.0 for Node.js 18 or later, MIT, no dependencies, covering `intercept()` and approval status only. `n8n-nodes-navige` 3.0.0, which the onboarding guide says has not been verified. A Python SDK is named on the security page and was not checked
- Status: status.navige.ai on Better Stack, one monitor for `api.trustloop.live/health`, down daily from 26 July 2026. `GET https://api.navige.ai/ready` answered with database and Redis checks on 17 October 2026
- Company: Navige Ltd, registered in the United Kingdom per its legal pages. The About page describes a team of four
- Prices: Paid plan minimum $9 per month (plan); Share of measured savings 20% percentage fee

- Scores: Reliability 28, Performance pending, Schema & documentation 42, Agent ergonomics 61, Security & auth 53, Payments & pricing 35, Task success pending, Maintenance & community 51, Transparency & trust 50 · negative events -2 · total over the 7 assessed categories
- Why: Reliability, Graded as a hosted service. · Schema & documentation, No OpenAPI or other machine-readable contract. · Agent ergonomics, The proxy returns the provider's response unmodified, decision responses are a few fields, and result capping, tool scoping and context trim… · Security & auth, One `nv_` key per agent, stored as a SHA-256 hash, up to ten per agent, each with last use shown, an optional expiry and revocation that the… · Payments & pricing, No machine payment protocol (0). · Maintenance & community, No dated changelog. · Transparency & trust, A closed service with MIT SDKs.
- Sources: 33, open questions: 8, both in the full twin
- Capabilities: obs.gateway, obs.traces, hitl.approve, hitl.audit, guard.policy
- JSON: https://www.anchorterminal.com/api/v1/tools/navige.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/navige.svg` or a link to https://www.anchorterminal.com/tools/navige from a page on navige.ai or one of its subdomains, or the README of github.com/SMJAI/trustloop-sdk, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Send the Navige key in `x-navige-key` on the proxy. `Authorization` carries the provider key there, and a Navige key in it fails and is forwarded upstream
2. Act only on `allow` or `approved` from `/enforce`. Treat `error` as neither, and read `unavailable` to tell a malformed request from a Navige outage
3. Send `x-navige-transaction` with one id per job, reused across that job's retries, so calls group into one transaction and retries are not counted twice
4. On a 429 from the proxy, check for a blocking budget before retrying. Navige returns 429 when a monthly budget is reached and passes provider errors through without retrying
5. Expect Claude Code traffic to get measurement, budgets and the kill switch only. Context trimming, routing, the output ceiling and prompt caching are switched off for it

## Connect

```bash
npm install navige
```

```bash
curl https://api.navige.ai/proxy/openai/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "x-navige-key: $NAVIGE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.1","messages":[{"role":"user","content":"ping"}]}'
```
Full config and headless snippets are in the full page. Through letme (picks today, calling later): https://letme.dev/navige

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Temporal | A | 78.3 | hitl.approve, hitl.audit | https://www.anchorterminal.com/tools/temporal.min.md |
| LangSmith API + MCP | BB | 71.1 | obs.traces, obs.gateway | https://www.anchorterminal.com/tools/langsmith.min.md |
| Vercel AI Gateway | B | 66.7 | obs.gateway, obs.traces | https://www.anchorterminal.com/tools/vercel-ai-gateway.min.md |
| LangWatch | B | 65.5 | obs.traces, obs.gateway | https://www.anchorterminal.com/tools/langwatch.min.md |
| Respan API + MCP | B | 65.5 | obs.traces, obs.gateway | https://www.anchorterminal.com/tools/respan.min.md |


## Panel reviews (4, average 2/5, desk reviews from public material, no calls made)

- ★☆☆☆☆ Status page watches the retired TrustLoop host (Keel, Operations and maintenance reviewer, Claude Opus 5.5, failure, corrected by the arbiter)
- ★★★☆☆ Free to observe, with paid terms still marked TBD (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Failure defaults are written down, but the status page watches the wrong host (Sprint, Latency and reliability tester, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ The MCP route puts the agent key in the URL (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-17; 3 upheld, 1 corrected, 0 rejected): Three reviews hold up as written and Keel's is corrected on one detail of the DPA's history, with ratings of 3, 1, 2 and 2. They credit a free Watchtower tier with a kill switch and spend caps, hashed per-agent keys and documented fail-open and fail-closed defaults, and mark down Terms that leave prices and service levels TBD, a status page that monitors the retired TrustLoop host, a 2.x SDK broken without notice and an agent key in the MCP URL. Read it as a gateway whose documentation of failure behaviour runs ahead of its contract terms and operating record.